video_reading
sleap_io.io.video_reading
¶
Backends for reading videos.
Classes:
| Name | Description |
|---|---|
CropVideoBackend |
Virtual, axis-aligned, on-read crop of an inner :class: |
HDF5Video |
Video backend for reading videos stored in HDF5 files. |
ImageVideo |
Video backend for reading videos stored as image files. |
MediaVideo |
Video backend for reading videos stored as common media files. |
TiffVideo |
Video backend for reading multi-page TIFF stacks. |
VideoBackend |
Base class for video backends. |
Functions:
| Name | Description |
|---|---|
crop_frame |
Crop a frame to the specified region. |
crop_points |
Adjust point coordinates for a crop transformation. |
get_available_image_backends |
Get list of available image backend plugins. |
get_available_video_backends |
Get list of available video backend plugins. |
get_default_image_plugin |
Get the current default image plugin. |
get_default_video_plugin |
Get the current default video plugin. |
get_installation_instructions |
Get installation instructions for backend plugins. |
normalize_image_plugin_name |
Normalize image plugin names to standard format. |
normalize_plugin_name |
Normalize plugin names to standard format. |
set_default_image_plugin |
Set the default image plugin for encoding/decoding embedded images. |
set_default_video_plugin |
Set the default video plugin for all subsequently loaded videos. |
uncrop_points |
Map crop-local point coordinates back to source coordinates. |
Attributes:
| Name | Type | Description |
|---|---|---|
__annotations__ |
dict() -> new empty dictionary |
|
__cached__ |
str(object='') -> str |
|
__doc__ |
str(object='') -> str |
|
__file__ |
str(object='') -> str |
|
__name__ |
str(object='') -> str |
|
__package__ |
str(object='') -> str |
__annotations__ = {'_default_video_plugin': 'str | None', '_default_image_plugin': 'str | None'}
module-attribute
¶
dict() -> new empty dictionary dict(mapping) -> new dictionary initialized from a mapping object's (key, value) pairs dict(iterable) -> new dictionary initialized as if via: d = {} for k, v in iterable: d[k] = v dict(**kwargs) -> new dictionary initialized with the name=value pairs in the keyword argument list. For example: dict(one=1, two=2)
__cached__ = '/home/runner/work/sleap-io/sleap-io/sleap_io/io/__pycache__/video_reading.cpython-313.pyc'
module-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__doc__ = 'Backends for reading videos.'
module-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__file__ = '/home/runner/work/sleap-io/sleap-io/sleap_io/io/video_reading.py'
module-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__name__ = 'sleap_io.io.video_reading'
module-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__package__ = 'sleap_io.io'
module-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
CropVideoBackend
¶
Bases: sleap_io.io.video_reading.VideoBackend
Virtual, axis-aligned, on-read crop of an inner :class:VideoBackend.
Wraps an inner backend and reports a cropped (F, h, w, c) view. Frames are
decoded by the inner backend, then cropped/padded by
:func:sleap_io.transform.frame.crop_frame (byte-identical), so no pixels are
copied or re-encoded on disk. Frame count is unchanged (a crop is spatial, not
temporal). Reports the cropped shape/grayscale and is a drop-in substitute
anywhere a VideoBackend is accepted.
Equality is by object identity (eq=False): inherited cache fields
(_cached_shape, _open_reader, ...) would otherwise pollute value
equality, and dedup never relies on backend == (it uses an explicit crop
key). Always construct via :meth:wrap (never the raw constructor) so the
"inner is never a crop" invariant and fill-aware flatten hold by construction.
Attributes:
| Name | Type | Description |
|---|---|---|
inner |
The wrapped source backend. Decodes full frames; this wrapper crops
its output. Invariant: |
|
crop |
Crop region |
|
fill |
Fill value for out-of-bounds regions, forwarded to |
|
owns_inner |
Whether this wrapper owns the inner backend's decode handle. If
|
|
filename |
Derived from |
Methods:
| Name | Description |
|---|---|
__attrs_post_init__ |
Derive |
__init__ |
Method generated by attrs for class CropVideoBackend. |
__repr__ |
Method generated by attrs for class CropVideoBackend. |
__setattr__ |
Method generated by attrs for class CropVideoBackend. |
close |
Release this wrapper's handle and the inner's, if owned. |
detect_grayscale |
Resolve grayscale from the inner, ignoring any passed cropped image. |
has_frame |
Check if a frame index is contained in the video (delegates to inner). |
read_test_frame |
Read the inner backend's UNCROPPED test frame (frame_map-safe). |
to_crop_coords |
Map source-frame |
to_source_coords |
Map cropped-frame |
wrap |
Wrap |
Source code in sleap_io/io/video_reading.py
@attrs.define(eq=False)
class CropVideoBackend(VideoBackend):
"""Virtual, axis-aligned, on-read crop of an inner :class:`VideoBackend`.
Wraps an inner backend and reports a cropped ``(F, h, w, c)`` view. Frames are
decoded by the inner backend, then cropped/padded by
:func:`sleap_io.transform.frame.crop_frame` (byte-identical), so no pixels are
copied or re-encoded on disk. Frame count is unchanged (a crop is spatial, not
temporal). Reports the cropped shape/grayscale and is a drop-in substitute
anywhere a ``VideoBackend`` is accepted.
Equality is by object identity (``eq=False``): inherited cache fields
(``_cached_shape``, ``_open_reader``, ...) would otherwise pollute value
equality, and dedup never relies on backend ``==`` (it uses an explicit crop
key). Always construct via :meth:`wrap` (never the raw constructor) so the
"inner is never a crop" invariant and fill-aware flatten hold by construction.
Attributes:
inner: The wrapped source backend. Decodes full frames; this wrapper crops
its output. Invariant: ``inner`` is never itself a ``CropVideoBackend``
(enforced by :meth:`wrap`).
crop: Crop region ``(x1, y1, x2, y2)``, ``x2``/``y2`` exclusive. May be
negative or exceed source bounds; out-of-bounds regions are pad-filled.
Stored as a tuple of ints.
fill: Fill value for out-of-bounds regions, forwarded to ``crop_frame``.
owns_inner: Whether this wrapper owns the inner backend's decode handle. If
``True`` (the default), :meth:`close` cascades to ``inner.close()``; if
``False`` (a shared-decode mosaic tile), it does not, so closing one
tile does not tear down siblings sharing the inner.
filename: Derived from ``inner.filename`` in post-init; NOT a constructor
argument.
"""
filename: str | Path | list[str] | list[Path] = attrs.field(
init=False, default=None
)
inner: VideoBackend = attrs.field(kw_only=True)
crop: tuple[int, int, int, int] = attrs.field(
kw_only=True, converter=lambda c: tuple(int(v) for v in c)
)
fill: int | tuple[int, ...] = attrs.field(
default=0,
kw_only=True,
converter=lambda f: (
tuple(int(v) for v in f) if isinstance(f, (list, tuple)) else f
),
)
owns_inner: bool = attrs.field(default=True, kw_only=True)
def __attrs_post_init__(self) -> None:
"""Derive ``filename`` from the inner; inherit resolved grayscale/fps lazily."""
object.__setattr__(self, "filename", getattr(self.inner, "filename", None))
# Inherit only an ALREADY-resolved inner grayscale; never force a decode at
# construction (keeps crop construction lazy). When unresolved, grayscale is
# resolved on first real access via the overridden ``detect_grayscale`` /
# ``img_shape``, always on the inner's FULL frame (so a degenerate crop never
# skews detection).
if self.grayscale is None and self.inner.grayscale is not None:
object.__setattr__(self, "grayscale", self.inner.grayscale)
# Carry fps so a closed/non-MediaVideo inner still reports fps.
if self._fps is None:
object.__setattr__(self, "_fps", getattr(self.inner, "_fps", None))
@classmethod
def wrap(
cls,
inner: VideoBackend,
crop: tuple[int, int, int, int],
fill: int | tuple[int, ...] = 0,
owns_inner: bool = True,
) -> "CropVideoBackend":
"""Wrap ``inner`` in a crop view, flattening crop-of-crop when safe.
Flattens (composes into a single wrapper) only when ``inner`` is itself a
``CropVideoBackend``, the fills agree, AND the outer crop lies fully within
the inner cropped frame ``[0, iw] x [0, ih]`` (``iw = ix2 - ix1``,
``ih = iy2 - iy1``). Otherwise it nests, preserving byte-parity:
- Different fills: the inner crop's materialized pad of value ``inner.fill``
would be silently replaced after a flatten.
- Outer crop exceeds the inner frame: a flatten would read real source
pixels where the nested view pads with ``fill``.
The flatten composition law expresses the outer rect in source coordinates:
``(ix1 + ox1, iy1 + oy1, ix1 + ox2, iy1 + oy2)``. A flattened ``inner`` is
always unwrapped to ``inner.inner`` so the "inner is never a crop"
invariant holds.
Args:
inner: The backend to wrap (may itself be a ``CropVideoBackend``).
crop: Outer crop region ``(x1, y1, x2, y2)``, expressed in the inner
(possibly already-cropped) frame.
fill: Fill value for out-of-bounds regions.
owns_inner: Whether the returned wrapper owns the inner's decode handle
(see the class ``owns_inner`` attribute).
Returns:
A ``CropVideoBackend`` whose ``inner`` is never a crop.
"""
# Normalize fills (a list and the equivalent tuple must compare equal so a
# crop reloaded from /video_crops still flattens against its tuple fill).
def _norm(f):
return tuple(int(v) for v in f) if isinstance(f, (list, tuple)) else f
if isinstance(inner, CropVideoBackend) and _norm(inner.fill) == _norm(fill):
ix1, iy1, ix2, iy2 = inner.crop
ox1, oy1, ox2, oy2 = (int(v) for v in crop)
iw, ih = ix2 - ix1, iy2 - iy1
if 0 <= ox1 and 0 <= oy1 and ox2 <= iw and oy2 <= ih:
crop = (ix1 + ox1, iy1 + oy1, ix1 + ox2, iy1 + oy2)
inner = inner.inner # invariant: inner.inner is never a crop
return cls(inner=inner, crop=crop, fill=fill, owns_inner=owns_inner)
@property
def num_frames(self) -> int:
"""Number of frames in the video (identity: a crop is spatial)."""
return self.inner.num_frames
@property
def img_shape(self) -> tuple[int, int, int]:
"""Shape of a single cropped frame as ``(height, width, channels)``.
Mirrors the inner's channel policy without forcing a decode here: when the
wrapper's grayscale is unresolved (``None``), inherit the inner's channel
count (``inner.img_shape`` resolves it from HDF5 attrs or a single detect),
so the cropped channels equal the source and stay stable across an SLP
round-trip. An explicit ``grayscale=True`` still collapses to 1 channel.
"""
x1, y1, x2, y2 = self.crop
c = 1 if self.grayscale else self.inner.img_shape[2]
return int(y2 - y1), int(x2 - x1), int(c)
@property
def dataset(self) -> object | None:
"""Inner backend's dataset name (delegated; ``None`` if absent)."""
return getattr(self.inner, "dataset", None)
@property
def input_format(self) -> object | None:
"""Inner backend's input format (delegated; ``None`` if absent)."""
return getattr(self.inner, "input_format", None)
def has_frame(self, frame_idx: int) -> bool:
"""Check if a frame index is contained in the video (delegates to inner).
Args:
frame_idx: Index of frame to check.
Returns:
``True`` if the inner backend contains the index, else ``False``.
"""
return self.inner.has_frame(frame_idx)
def read_test_frame(self) -> np.ndarray:
"""Read the inner backend's UNCROPPED test frame (frame_map-safe)."""
return self.inner.read_test_frame()
def detect_grayscale(self, test_img: np.ndarray | None = None) -> bool:
"""Resolve grayscale from the inner, ignoring any passed cropped image.
Args:
test_img: Ignored; grayscale is always resolved on the inner's full
frame so a degenerate (e.g. 1px-wide) crop never breaks detection.
Returns:
Whether the video is grayscale (also cached on ``self.grayscale``).
"""
gs = self.inner.grayscale
if gs is None:
gs = self.inner.detect_grayscale()
self.grayscale = gs
self._cached_shape = None
return gs
def close(self) -> None:
"""Release this wrapper's handle and the inner's, if owned.
Always releases the wrapper's own (unused) cached reader via
:meth:`VideoBackend.close`, then cascades to ``inner.close()`` only when
``owns_inner`` (a shared-decode mosaic tile leaves the shared inner open).
"""
super().close()
if self.owns_inner:
self.inner.close()
def _source_frame(self, frame_idx: int) -> tuple[np.ndarray, bool]:
"""Return ``(frame, already_cropped)`` for a single frame.
Tries the inner's HDF5 crop pushdown first (returns the final cropped
frame); otherwise delegates to ``inner._read_frame`` (NOT ``get_frame``, so
the base ``get_frame`` applies the grayscale slice exactly once to our
cropped result).
Args:
frame_idx: Index of frame to read.
Returns:
A tuple of the frame and a flag indicating whether it is already
cropped (pushdown) or still a full source frame.
"""
read_crop = getattr(self.inner, "read_crop", None)
if read_crop is not None:
out = read_crop(frame_idx, self.crop, self.fill)
if out is not None:
return out, True
return self.inner._read_frame(frame_idx), False
def _apply_view(self, frame: np.ndarray, frame_idx: int) -> np.ndarray:
"""Crop/pad a full source frame into the cropped view.
Args:
frame: A full source frame ``(H, W, C)``.
frame_idx: Reserved for a future per-frame-varying window; unused.
Returns:
The cropped frame, materialized as a contiguous owned array.
"""
return np.ascontiguousarray(crop_frame(frame, self.crop, fill=self.fill))
def _read_frame(self, frame_idx: int) -> np.ndarray:
"""Read a single cropped frame.
Args:
frame_idx: Index of frame to read.
Returns:
The cropped frame ``(crop_h, crop_w, C)``.
"""
src, already_cropped = self._source_frame(frame_idx)
return src if already_cropped else self._apply_view(src, frame_idx)
def _read_frames(self, frame_inds: list) -> np.ndarray:
"""Read a list of cropped frames.
Tries the inner's batched HDF5 crop pushdown first; otherwise reads full
frames from the inner and crops them. The fully-in-bounds case uses a basic
slice then ``.copy()`` to defuse aliasing into the inner's reused decode
buffer; otherwise it pads per-frame via ``crop_frame``.
Args:
frame_inds: List of frame indices to read.
Returns:
Cropped frames ``(N, crop_h, crop_w, C)``.
"""
read_crops = getattr(self.inner, "read_crops", None)
if read_crops is not None:
out = read_crops(frame_inds, self.crop, self.fill)
if out is not None:
return out
imgs = self.inner._read_frames(frame_inds)
x1, y1, x2, y2 = self.crop
h, w = imgs.shape[1], imgs.shape[2]
if 0 <= x1 and 0 <= y1 and x2 <= w and y2 <= h:
return imgs[:, y1:y2, x1:x2].copy()
return np.stack(
[crop_frame(im, self.crop, fill=self.fill) for im in imgs], axis=0
)
def to_crop_coords(self, points: np.ndarray) -> np.ndarray:
"""Map source-frame ``(x, y)`` coordinates into the cropped frame.
Args:
points: Coordinate array of shape ``(..., 2)``. NaN values are
preserved.
Returns:
Coordinates translated by ``-(x1, y1)`` (copy-based, NaN-preserving).
"""
return crop_points(points, self.crop)
def to_source_coords(self, points: np.ndarray) -> np.ndarray:
"""Map cropped-frame ``(x, y)`` coordinates back to source coordinates.
Inverse of :meth:`to_crop_coords`.
Args:
points: Coordinate array of shape ``(..., 2)``. NaN values are
preserved.
Returns:
Coordinates translated by ``+(x1, y1)`` (copy-based, NaN-preserving).
"""
return uncrop_points(points, self.crop)
__annotations__ = {'filename': 'str | Path | list[str] | list[Path]', 'inner': 'VideoBackend', 'crop': 'tuple[int, int, int, int]', 'fill': 'int | tuple[int, ...]', 'owns_inner': 'bool'}
class-attribute
¶
dict() -> new empty dictionary dict(mapping) -> new dictionary initialized from a mapping object's (key, value) pairs dict(iterable) -> new dictionary initialized as if via: d = {} for k, v in iterable: d[k] = v dict(**kwargs) -> new dictionary initialized with the name=value pairs in the keyword argument list. For example: dict(one=1, two=2)
__attrs_own_setattr__ = True
class-attribute
¶
Returns True when the argument is true, False otherwise. The builtins True and False are the only two instances of the class bool. The class bool is a subclass of the class int, and cannot be subclassed.
__attrs_props__ = ClassProps(is_exception=False, is_slotted=True, has_weakref_slot=True, is_frozen=False, kw_only=<KeywordOnly.NO: 'no'>, collected_fields_by_mro=True, added_init=True, added_repr=True, added_eq=False, added_ordering=False, hashability=<Hashability.LEAVE_ALONE: 'leave_alone'>, added_match_args=True, added_str=False, added_pickling=True, on_setattr_hook=<function pipe.<locals>.wrapped_pipe at 0x7f08471ce840>, field_transformer=None)
class-attribute
¶
Effective class properties as derived from parameters to attr.s() or
define() decorators.
This is the same data structure that attrs uses internally to decide how to construct the final class.
Warning:
This feature is currently **experimental** and is not covered by our
strict backwards-compatibility guarantees.
Attributes:
| Name | Type | Description |
|---|---|---|
is_exception |
bool
|
Whether the class is treated as an exception class. |
is_slotted |
bool
|
Whether the class is |
has_weakref_slot |
bool
|
Whether the class has a slot for weak references. |
is_frozen |
bool
|
Whether the class is frozen. |
kw_only |
KeywordOnly
|
Whether / how the class enforces keyword-only arguments on the
|
collected_fields_by_mro |
bool
|
Whether the class fields were collected by method resolution order.
That is, correctly but unlike |
added_init |
bool
|
Whether the class has an attrs-generated |
added_repr |
bool
|
Whether the class has an attrs-generated |
added_eq |
bool
|
Whether the class has attrs-generated equality methods. |
added_ordering |
bool
|
Whether the class has attrs-generated ordering methods. |
hashability |
Hashability
|
How |
added_match_args |
bool
|
Whether the class supports positional |
added_str |
bool
|
Whether the class has an attrs-generated |
added_pickling |
bool
|
Whether the class has attrs-generated |
on_setattr_hook |
Callable[[Any, Attribute[Any], Any], Any] | None
|
The class's |
field_transformer |
Callable[[Attribute[Any]], Attribute[Any]] | None
|
The class's |
.. versionadded:: 25.4.0
__doc__ = 'Virtual, axis-aligned, on-read crop of an inner :class:`VideoBackend`.\n\nWraps an inner backend and reports a cropped ``(F, h, w, c)`` view. Frames are\ndecoded by the inner backend, then cropped/padded by\n:func:`sleap_io.transform.frame.crop_frame` (byte-identical), so no pixels are\ncopied or re-encoded on disk. Frame count is unchanged (a crop is spatial, not\ntemporal). Reports the cropped shape/grayscale and is a drop-in substitute\nanywhere a ``VideoBackend`` is accepted.\n\nEquality is by object identity (``eq=False``): inherited cache fields\n(``_cached_shape``, ``_open_reader``, ...) would otherwise pollute value\nequality, and dedup never relies on backend ``==`` (it uses an explicit crop\nkey). Always construct via :meth:`wrap` (never the raw constructor) so the\n"inner is never a crop" invariant and fill-aware flatten hold by construction.\n\nAttributes:\n inner: The wrapped source backend. Decodes full frames; this wrapper crops\n its output. Invariant: ``inner`` is never itself a ``CropVideoBackend``\n (enforced by :meth:`wrap`).\n crop: Crop region ``(x1, y1, x2, y2)``, ``x2``/``y2`` exclusive. May be\n negative or exceed source bounds; out-of-bounds regions are pad-filled.\n Stored as a tuple of ints.\n fill: Fill value for out-of-bounds regions, forwarded to ``crop_frame``.\n owns_inner: Whether this wrapper owns the inner backend\'s decode handle. If\n ``True`` (the default), :meth:`close` cascades to ``inner.close()``; if\n ``False`` (a shared-decode mosaic tile), it does not, so closing one\n tile does not tear down siblings sharing the inner.\n filename: Derived from ``inner.filename`` in post-init; NOT a constructor\n argument.\n'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__firstlineno__ = 2282
class-attribute
¶
int([x]) -> integer int(x, base=10) -> integer
Convert a number or string to an integer, or return 0 if no arguments are given. If x is a number, return x.int(). For floating-point numbers, this truncates towards zero.
If x is not a number or if base is given, then x must be a string, bytes, or bytearray instance representing an integer literal in the given base. The literal can be preceded by '+' or '-' and be surrounded by whitespace. The base defaults to 10. Valid bases are 0 and 2-36. Base 0 means to interpret the base from the string as an integer iteral.
int('0b100', base=0) 4
__match_args__ = ('grayscale', 'keep_open', '_cached_shape', '_open_reader', '_fps')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__module__ = 'sleap_io.io.video_reading'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__slots__ = ('inner', 'crop', 'fill', 'owns_inner')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__static_attributes__ = ('_cached_shape', 'grayscale')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
dataset
property
¶
Inner backend's dataset name (delegated; None if absent).
img_shape
property
¶
Shape of a single cropped frame as (height, width, channels).
Mirrors the inner's channel policy without forcing a decode here: when the
wrapper's grayscale is unresolved (None), inherit the inner's channel
count (inner.img_shape resolves it from HDF5 attrs or a single detect),
so the cropped channels equal the source and stay stable across an SLP
round-trip. An explicit grayscale=True still collapses to 1 channel.
input_format
property
¶
Inner backend's input format (delegated; None if absent).
num_frames
property
¶
Number of frames in the video (identity: a crop is spatial).
__attrs_post_init__()
¶
Derive filename from the inner; inherit resolved grayscale/fps lazily.
Source code in sleap_io/io/video_reading.py
def __attrs_post_init__(self) -> None:
"""Derive ``filename`` from the inner; inherit resolved grayscale/fps lazily."""
object.__setattr__(self, "filename", getattr(self.inner, "filename", None))
# Inherit only an ALREADY-resolved inner grayscale; never force a decode at
# construction (keeps crop construction lazy). When unresolved, grayscale is
# resolved on first real access via the overridden ``detect_grayscale`` /
# ``img_shape``, always on the inner's FULL frame (so a degenerate crop never
# skews detection).
if self.grayscale is None and self.inner.grayscale is not None:
object.__setattr__(self, "grayscale", self.inner.grayscale)
# Carry fps so a closed/non-MediaVideo inner still reports fps.
if self._fps is None:
object.__setattr__(self, "_fps", getattr(self.inner, "_fps", None))
__init__(grayscale=None, keep_open=True, cached_shape=None, open_reader=None, fps=None, *, inner, crop, fill=0, owns_inner=True)
¶
Method generated by attrs for class CropVideoBackend.
__repr__()
¶
Method generated by attrs for class CropVideoBackend.
__setattr__(name, val)
¶
Method generated by attrs for class CropVideoBackend.
Source code in sleap_io/io/video_reading.py
multiple threads are safe: although all reads share one cached fsspec
file-like (a single byte position), h5py serializes every HDF5 C-library
call under a global recursive lock (`h5py._objects.phil`), so the
seek+read pair a frame read performs is never interleaved across threads.
For true read *parallelism* (rather than just safety), construct
independent `Video`/`HDF5Video` instances per worker; each gets its own
fsspec file and block cache.
"""
close()
¶
Release this wrapper's handle and the inner's, if owned.
Always releases the wrapper's own (unused) cached reader via
:meth:VideoBackend.close, then cascades to inner.close() only when
owns_inner (a shared-decode mosaic tile leaves the shared inner open).
Source code in sleap_io/io/video_reading.py
def close(self) -> None:
"""Release this wrapper's handle and the inner's, if owned.
Always releases the wrapper's own (unused) cached reader via
:meth:`VideoBackend.close`, then cascades to ``inner.close()`` only when
``owns_inner`` (a shared-decode mosaic tile leaves the shared inner open).
"""
super().close()
if self.owns_inner:
self.inner.close()
detect_grayscale(test_img=None)
¶
Resolve grayscale from the inner, ignoring any passed cropped image.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
test_img
|
ndarray | None
|
Ignored; grayscale is always resolved on the inner's full frame so a degenerate (e.g. 1px-wide) crop never breaks detection. |
None
|
Returns:
| Type | Description |
|---|---|
bool
|
Whether the video is grayscale (also cached on |
Source code in sleap_io/io/video_reading.py
def detect_grayscale(self, test_img: np.ndarray | None = None) -> bool:
"""Resolve grayscale from the inner, ignoring any passed cropped image.
Args:
test_img: Ignored; grayscale is always resolved on the inner's full
frame so a degenerate (e.g. 1px-wide) crop never breaks detection.
Returns:
Whether the video is grayscale (also cached on ``self.grayscale``).
"""
gs = self.inner.grayscale
if gs is None:
gs = self.inner.detect_grayscale()
self.grayscale = gs
self._cached_shape = None
return gs
has_frame(frame_idx)
¶
Check if a frame index is contained in the video (delegates to inner).
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame_idx
|
int
|
Index of frame to check. |
required |
Returns:
| Type | Description |
|---|---|
bool
|
|
Source code in sleap_io/io/video_reading.py
read_test_frame()
¶
to_crop_coords(points)
¶
Map source-frame (x, y) coordinates into the cropped frame.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
points
|
ndarray
|
Coordinate array of shape |
required |
Returns:
| Type | Description |
|---|---|
ndarray
|
Coordinates translated by |
Source code in sleap_io/io/video_reading.py
def to_crop_coords(self, points: np.ndarray) -> np.ndarray:
"""Map source-frame ``(x, y)`` coordinates into the cropped frame.
Args:
points: Coordinate array of shape ``(..., 2)``. NaN values are
preserved.
Returns:
Coordinates translated by ``-(x1, y1)`` (copy-based, NaN-preserving).
"""
return crop_points(points, self.crop)
to_source_coords(points)
¶
Map cropped-frame (x, y) coordinates back to source coordinates.
Inverse of :meth:to_crop_coords.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
points
|
ndarray
|
Coordinate array of shape |
required |
Returns:
| Type | Description |
|---|---|
ndarray
|
Coordinates translated by |
Source code in sleap_io/io/video_reading.py
def to_source_coords(self, points: np.ndarray) -> np.ndarray:
"""Map cropped-frame ``(x, y)`` coordinates back to source coordinates.
Inverse of :meth:`to_crop_coords`.
Args:
points: Coordinate array of shape ``(..., 2)``. NaN values are
preserved.
Returns:
Coordinates translated by ``+(x1, y1)`` (copy-based, NaN-preserving).
"""
return uncrop_points(points, self.crop)
wrap(inner, crop, fill=0, owns_inner=True)
classmethod
¶
Wrap inner in a crop view, flattening crop-of-crop when safe.
Flattens (composes into a single wrapper) only when inner is itself a
CropVideoBackend, the fills agree, AND the outer crop lies fully within
the inner cropped frame [0, iw] x [0, ih] (iw = ix2 - ix1,
ih = iy2 - iy1). Otherwise it nests, preserving byte-parity:
- Different fills: the inner crop's materialized pad of value
inner.fillwould be silently replaced after a flatten. - Outer crop exceeds the inner frame: a flatten would read real source
pixels where the nested view pads with
fill.
The flatten composition law expresses the outer rect in source coordinates:
(ix1 + ox1, iy1 + oy1, ix1 + ox2, iy1 + oy2). A flattened inner is
always unwrapped to inner.inner so the "inner is never a crop"
invariant holds.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
inner
|
VideoBackend
|
The backend to wrap (may itself be a |
required |
crop
|
tuple[int, int, int, int]
|
Outer crop region |
required |
fill
|
int | tuple[int, ...]
|
Fill value for out-of-bounds regions. |
0
|
owns_inner
|
bool
|
Whether the returned wrapper owns the inner's decode handle
(see the class |
True
|
Returns:
| Type | Description |
|---|---|
CropVideoBackend
|
A |
Source code in sleap_io/io/video_reading.py
@classmethod
def wrap(
cls,
inner: VideoBackend,
crop: tuple[int, int, int, int],
fill: int | tuple[int, ...] = 0,
owns_inner: bool = True,
) -> "CropVideoBackend":
"""Wrap ``inner`` in a crop view, flattening crop-of-crop when safe.
Flattens (composes into a single wrapper) only when ``inner`` is itself a
``CropVideoBackend``, the fills agree, AND the outer crop lies fully within
the inner cropped frame ``[0, iw] x [0, ih]`` (``iw = ix2 - ix1``,
``ih = iy2 - iy1``). Otherwise it nests, preserving byte-parity:
- Different fills: the inner crop's materialized pad of value ``inner.fill``
would be silently replaced after a flatten.
- Outer crop exceeds the inner frame: a flatten would read real source
pixels where the nested view pads with ``fill``.
The flatten composition law expresses the outer rect in source coordinates:
``(ix1 + ox1, iy1 + oy1, ix1 + ox2, iy1 + oy2)``. A flattened ``inner`` is
always unwrapped to ``inner.inner`` so the "inner is never a crop"
invariant holds.
Args:
inner: The backend to wrap (may itself be a ``CropVideoBackend``).
crop: Outer crop region ``(x1, y1, x2, y2)``, expressed in the inner
(possibly already-cropped) frame.
fill: Fill value for out-of-bounds regions.
owns_inner: Whether the returned wrapper owns the inner's decode handle
(see the class ``owns_inner`` attribute).
Returns:
A ``CropVideoBackend`` whose ``inner`` is never a crop.
"""
# Normalize fills (a list and the equivalent tuple must compare equal so a
# crop reloaded from /video_crops still flattens against its tuple fill).
def _norm(f):
return tuple(int(v) for v in f) if isinstance(f, (list, tuple)) else f
if isinstance(inner, CropVideoBackend) and _norm(inner.fill) == _norm(fill):
ix1, iy1, ix2, iy2 = inner.crop
ox1, oy1, ox2, oy2 = (int(v) for v in crop)
iw, ih = ix2 - ix1, iy2 - iy1
if 0 <= ox1 and 0 <= oy1 and ox2 <= iw and oy2 <= ih:
crop = (ix1 + ox1, iy1 + oy1, ix1 + ox2, iy1 + oy2)
inner = inner.inner # invariant: inner.inner is never a crop
return cls(inner=inner, crop=crop, fill=fill, owns_inner=owns_inner)
HDF5Video
¶
Bases: sleap_io.io.video_reading.VideoBackend
Video backend for reading videos stored in HDF5 files.
This backend supports reading videos stored in HDF5 files, both in rank-4 datasets as well as in datasets with lists of binary-encoded images.
Embedded image datasets are used in SLEAP when exporting package files (.pkg.slp)
with videos embedded in them. This is useful for bundling training or inference data
without having to worry about the videos (or frame images) being moved or deleted.
It is expected that these types of datasets will be in a Group with a int8
variable length dataset called "video". This dataset must also contain an
attribute called "format" with a string describing the image format (e.g., "png" or
"jpg") which will be used to decode it appropriately.
If a frame_numbers dataset is present in the group, it will be used to map from
source video frames to the frames in the dataset. This is useful to preserve frame
indexing when exporting a subset of frames in the video. It will also be used to
populate frame_map and source_inds attributes.
Attributes:
| Name | Type | Description |
|---|---|---|
filename |
Path to HDF5 file (.h5, .hdf5 or .slp). |
|
grayscale |
Whether to force grayscale. If None, autodetect on first frame load. |
|
keep_open |
Whether to keep the video reader open between calls to read frames. If False, will close the reader after each call. If True (the default), it will keep the reader open and cache it for subsequent calls which may enhance the performance of reading multiple frames. |
|
dataset |
Name of dataset to read from. If |
|
input_format |
Format of the data in the dataset. One of "channels_last" (the
default) in |
|
frame_map |
Mapping from frame indices to indices in the dataset. This is used to translate between the frame indices of the images within their source video and the indices of the images in the dataset. This is only used when reading embedded image datasets. |
|
source_filename |
Path to the source video file. This is metadata and only used when reading embedded image datasets. |
|
source_inds |
Indices of the frames in the source video file. This is metadata and only used when reading embedded image datasets. |
|
image_format |
Format of the images in the embedded dataset. This is metadata and only used when reading embedded image datasets. |
|
channel_order |
Channel order of embedded images, either "RGB" or "BGR". This is used to ensure consistent color channel ordering when decoding embedded images. If the encoding and decoding plugins have different channel orders, the channels will be automatically flipped during decoding. |
|
plugin |
Plugin to use for decoding embedded images. One of "opencv" or "FFMPEG". If None, uses the global default or auto-detects based on available packages. Note that "pyav" is automatically mapped to "FFMPEG" since PyAV doesn't support image decoding. |
Notes
Concurrent reads of a single remote (URL-backed) HDF5Video from
multiple threads are safe: although all reads share one cached fsspec
file-like (a single byte position), h5py serializes every HDF5 C-library
call under a global recursive lock (h5py._objects.phil), so the
seek+read pair a frame read performs is never interleaved across threads.
For true read parallelism (rather than just safety), construct
independent Video/HDF5Video instances per worker; each gets its own
fsspec file and block cache.
Methods:
| Name | Description |
|---|---|
__attrs_post_init__ |
Auto-detect dataset and frame map heuristically. |
__eq__ |
Method generated by attrs for class HDF5Video. |
__getstate__ |
Return state for pickling/deepcopy, dropping unpicklable handles. |
__init__ |
Method generated by attrs for class HDF5Video. |
__repr__ |
Method generated by attrs for class HDF5Video. |
__setattr__ |
Method generated by attrs for class HDF5Video. |
close |
Release the cached HDF5 reader and the cached fsspec URL file-like. |
decode_embedded |
Decode an embedded image string into a numpy array. |
get_frame_raw_bytes |
Get raw encoded bytes for a frame without decoding. |
has_frame |
Check if a frame index is contained in the video. |
read_crop |
Read a spatial hyperslab of a frame, padded to the crop shape. |
read_crops |
Batched :meth: |
read_test_frame |
Read a single frame from the video to test for grayscale. |
Source code in sleap_io/io/video_reading.py
@attrs.define
class HDF5Video(VideoBackend):
"""Video backend for reading videos stored in HDF5 files.
This backend supports reading videos stored in HDF5 files, both in rank-4 datasets
as well as in datasets with lists of binary-encoded images.
Embedded image datasets are used in SLEAP when exporting package files (`.pkg.slp`)
with videos embedded in them. This is useful for bundling training or inference data
without having to worry about the videos (or frame images) being moved or deleted.
It is expected that these types of datasets will be in a `Group` with a `int8`
variable length dataset called `"video"`. This dataset must also contain an
attribute called "format" with a string describing the image format (e.g., "png" or
"jpg") which will be used to decode it appropriately.
If a `frame_numbers` dataset is present in the group, it will be used to map from
source video frames to the frames in the dataset. This is useful to preserve frame
indexing when exporting a subset of frames in the video. It will also be used to
populate `frame_map` and `source_inds` attributes.
Attributes:
filename: Path to HDF5 file (.h5, .hdf5 or .slp).
grayscale: Whether to force grayscale. If None, autodetect on first frame load.
keep_open: Whether to keep the video reader open between calls to read frames.
If False, will close the reader after each call. If True (the default), it
will keep the reader open and cache it for subsequent calls which may
enhance the performance of reading multiple frames.
dataset: Name of dataset to read from. If `None`, will try to find a rank-4
dataset by iterating through datasets in the file. If specifying an embedded
dataset, this can be the group containing a "video" dataset or the dataset
itself (e.g., "video0" or "video0/video").
input_format: Format of the data in the dataset. One of "channels_last" (the
default) in `(frames, height, width, channels)` order or "channels_first" in
`(frames, channels, width, height)` order. Embedded datasets should use the
"channels_last" format.
frame_map: Mapping from frame indices to indices in the dataset. This is used to
translate between the frame indices of the images within their source video
and the indices of the images in the dataset. This is only used when reading
embedded image datasets.
source_filename: Path to the source video file. This is metadata and only used
when reading embedded image datasets.
source_inds: Indices of the frames in the source video file. This is metadata
and only used when reading embedded image datasets.
image_format: Format of the images in the embedded dataset. This is metadata and
only used when reading embedded image datasets.
channel_order: Channel order of embedded images, either "RGB" or "BGR". This is
used to ensure consistent color channel ordering when decoding embedded
images. If the encoding and decoding plugins have different channel orders,
the channels will be automatically flipped during decoding.
plugin: Plugin to use for decoding embedded images. One of "opencv" or
"FFMPEG". If None, uses the global default or auto-detects based on
available packages. Note that "pyav" is automatically mapped to "FFMPEG"
since PyAV doesn't support image decoding.
Notes:
Concurrent reads of a single remote (URL-backed) `HDF5Video` from
multiple threads are safe: although all reads share one cached fsspec
file-like (a single byte position), h5py serializes every HDF5 C-library
call under a global recursive lock (`h5py._objects.phil`), so the
seek+read pair a frame read performs is never interleaved across threads.
For true read *parallelism* (rather than just safety), construct
independent `Video`/`HDF5Video` instances per worker; each gets its own
fsspec file and block cache.
"""
dataset: str | None = None
input_format: str = attrs.field(
default="channels_last",
validator=attrs.validators.in_(["channels_last", "channels_first"]),
)
frame_map: dict[int, int] = attrs.field(init=False, default=attrs.Factory(dict))
_can_push_crop_cached: bool | None = attrs.field(
init=False, default=None, repr=False, eq=False
)
source_filename: str | None = None
source_inds: np.ndarray | None = None
image_format: str = "hdf5"
channel_order: str = "RGB"
plugin: str | None = None
_url_file: object | None = attrs.field(
init=False, default=None, repr=False, eq=False
)
# ``_url_headers`` / ``_url_stream_mode`` are ``init=True`` (attrs derives the
# constructor aliases ``url_headers`` / ``url_stream_mode`` by stripping the
# leading underscore) so the metadata probe in ``__attrs_post_init__`` runs
# *authenticated*: an embedded ``pkg.slp`` over an auth-gated URL would
# otherwise probe with no headers and silently lose the embedded-image
# metadata. They remain ``repr=False, eq=False`` and, because the attribute
# names keep the leading underscore, the name-based ``__getstate__`` pickle
# contract is unchanged.
_url_headers: dict[str, str] | None = attrs.field(
default=None, repr=False, eq=False
)
_url_stream_mode: str = attrs.field(default="blockcache", repr=False, eq=False)
EXTS = ("h5", "hdf5", "slp")
def _open_h5(self) -> h5py.File:
"""Open the backing HDF5 file as an ``h5py.File`` in read mode.
For local paths this opens ``self.filename`` directly. For URLs it lazily
opens (and caches on ``self._url_file``) an fsspec-backed file-like object
via :func:`sleap_io.io._remote.open_url` and wraps it with ``h5py``.
Returns:
An open ``h5py.File`` handle. The caller owns closing the returned
handle; the cached ``self._url_file`` is reused across reads and
dropped on pickling.
"""
from sleap_io.io import _remote
if _remote._is_url(self.filename):
if self._url_file is None:
self._url_file = _remote.open_url(
self.filename,
headers=self._url_headers,
stream_mode=self._url_stream_mode,
)
return h5py.File(self._url_file, "r")
return h5py.File(self.filename, "r")
def _close_url_file(self) -> None:
"""Close and drop the cached fsspec URL file-like, if any (idempotent).
A no-op for local files (where ``_url_file`` is never set) and when it
has already been closed/dropped.
"""
if self._url_file is None:
return
try:
self._url_file.close()
except Exception: # pragma: no cover - defensive: close() should not raise
pass
self._url_file = None
def _release_probe_url_file(self, preexisting: bool) -> None:
"""Close and drop ``self._url_file`` if a probe opened it.
Used by :meth:`__attrs_post_init__` so that a remote handle opened just
to sniff the dataset/frame-map does not leak and is not reused by later
(possibly authenticated) reads. A no-op for local files and when the
cached file-like already existed before the probe.
Args:
preexisting: Whether ``self._url_file`` was already set before the
probe opened the file (in which case it is left untouched).
"""
if preexisting:
return
self._close_url_file()
def close(self) -> None:
"""Release the cached HDF5 reader and the cached fsspec URL file-like.
Extends :meth:`VideoBackend.close` (which drops the cached ``h5py.File``
reader) by also closing the fsspec-backed ``_url_file`` shared across
reads, which ``h5py.File.close()`` does not close on its own. Both are
lazily reopened on the next read, so this is safe to call between reads.
"""
super().close()
self._close_url_file()
def __getstate__(self) -> dict:
"""Return state for pickling/deepcopy, dropping unpicklable handles.
Extends :meth:`VideoBackend.__getstate__` to also drop the cached
fsspec-backed ``_url_file`` (reopened lazily by :meth:`_open_h5`).
"""
state = super().__getstate__()
state["_url_file"] = None
return state
def __attrs_post_init__(self):
"""Auto-detect dataset and frame map heuristically."""
# Check if the file accessible before applying heuristics.
# For URLs, track whether this probe opened the cached fsspec file-like
# so it can be released afterwards (it would otherwise leak the handle on
# an early return / exception, and a probe-time open may predate the
# final auth headers being applied).
url_file_preexisting = self._url_file is not None
try:
f = self._open_h5()
except OSError:
self._release_probe_url_file(url_file_preexisting)
return
try:
if self.dataset is None:
# Iterate through datasets to find a rank 4 array.
def find_movies(name, obj):
if isinstance(obj, h5py.Dataset) and obj.ndim == 4:
self.dataset = name
return True
f.visititems(find_movies)
if self.dataset is None:
# Iterate through datasets to find an embedded video dataset.
def find_embedded(name, obj):
if isinstance(obj, h5py.Dataset) and name.endswith("/video"):
self.dataset = name
return True
f.visititems(find_embedded)
if self.dataset is None:
# Couldn't find video datasets.
return
if isinstance(f[self.dataset], h5py.Group):
# If this is a group, assume it's an embedded video dataset.
if "video" in f[self.dataset]:
self.dataset = f"{self.dataset}/video"
if self.dataset.split("/")[-1] == "video":
# This may be an embedded video dataset. Check for frame map.
ds = f[self.dataset]
if "format" in ds.attrs:
self.image_format = ds.attrs["format"]
# Read channel_order, with backwards compatibility
if "channel_order" in ds.attrs:
self.channel_order = ds.attrs["channel_order"]
else:
# Backwards compatibility: Check format_id for older files
# Prior to format 1.4, embedded images were primarily encoded
# with OpenCV which uses BGR, so default to BGR for older
# formats
if "metadata" in f and "format_id" in f["metadata"].attrs:
format_id = f["metadata"].attrs["format_id"]
if format_id < 1.4:
self.channel_order = "BGR" # Legacy default
# If no format_id found, assume BGR (safest legacy default)
# since most embedded images before this change used OpenCV
if "frame_numbers" in ds.parent:
frame_numbers = ds.parent["frame_numbers"][:].astype(int)
self.frame_map = {
frame: idx for idx, frame in enumerate(frame_numbers)
}
self.source_inds = frame_numbers
if "source_video" in ds.parent:
source_grp = ds.parent["source_video"]
# Source metadata is normally in the "json" attribute, but
# oversized metadata (e.g. an image-sequence source with many
# thousands of filenames, exceeding HDF5's 64 KB attribute limit)
# is stored in a "json" *dataset* instead (see
# ``slp._write_source_video_json``). Read whichever is present so
# such packages remain openable -- otherwise the backend fails to
# open, ``Video.backend`` is left ``None``, and embedded frames
# cannot be read.
if "json" in source_grp:
source_json = source_grp["json"][()]
else:
source_json = source_grp.attrs["json"]
self.source_filename = json.loads(source_json)["backend"][
"filename"
]
# Read FPS from attributes if present
if "fps" in ds.attrs:
self._fps = float(ds.attrs["fps"])
elif "fps" in ds.parent.attrs:
self._fps = float(ds.parent.attrs["fps"])
finally:
f.close()
self._release_probe_url_file(url_file_preexisting)
# Set default plugin if not specified (use image plugin, not video plugin)
if self.plugin is None:
# Check image plugin default first (for embedded images)
if _default_image_plugin is not None:
self.plugin = _default_image_plugin
# Otherwise auto-detect (for embedded image decoding)
elif "cv2" in sys.modules:
self.plugin = "opencv"
else:
self.plugin = "imageio" # imageio fallback
@property
def num_frames(self) -> int:
"""Number of frames in the video."""
with self._open_h5() as f:
return f[self.dataset].shape[0]
@property
def img_shape(self) -> tuple[int, int, int]:
"""Shape of a single frame in the video as `(height, width, channels)`."""
with self._open_h5() as f:
ds = f[self.dataset]
img_shape = None
if "height" in ds.attrs:
# Try to get shape from the attributes.
img_shape = (
ds.attrs["height"],
ds.attrs["width"],
ds.attrs["channels"],
)
if img_shape[0] == 0 or img_shape[1] == 0:
# Invalidate the shape if the attributes are zero.
img_shape = None
if img_shape is None and self.image_format == "hdf5" and ds.ndim == 4:
# Use the dataset shape if just stored as a rank-4 array.
img_shape = ds.shape[1:]
if self.input_format == "channels_first":
img_shape = img_shape[::-1]
if img_shape is None:
# Fall back to reading a test frame.
return super().img_shape
return int(img_shape[0]), int(img_shape[1]), int(img_shape[2])
def read_test_frame(self) -> np.ndarray:
"""Read a single frame from the video to test for grayscale."""
if self.frame_map:
frame_idx = list(self.frame_map.keys())[0]
else:
frame_idx = 0
return self._read_frame(frame_idx)
@property
def has_embedded_images(self) -> bool:
"""Return True if the dataset contains embedded images."""
return self.image_format is not None and self.image_format != "hdf5"
@property
def embedded_frame_inds(self) -> list[int]:
"""Return the frame indices of the embedded images."""
return list(self.frame_map.keys())
def decode_embedded(self, img_string: np.ndarray) -> np.ndarray:
"""Decode an embedded image string into a numpy array.
Args:
img_string: Binary string of the image as a `int8` numpy vector with the
bytes as values corresponding to the format-encoded image.
Returns:
The decoded image as a numpy array of shape `(height, width, channels)`. If
a rank-2 image is decoded, it will be expanded such that channels will be 1.
This method does not apply grayscale conversion as per the `grayscale`
attribute. Use the `get_frame` or `get_frames` methods of the `VideoBackend`
to apply grayscale conversion rather than calling this function directly.
"""
# Decode based on plugin
if self.plugin == "opencv":
img = cv2.imdecode(img_string, cv2.IMREAD_UNCHANGED)
decoder_order = "BGR" # OpenCV decodes to BGR
else:
# Use imageio for FFMPEG or any other plugin
img = iio.imread(BytesIO(img_string), extension=f".{self.image_format}")
decoder_order = "RGB" # imageio decodes to RGB
if img.ndim == 2:
img = np.expand_dims(img, axis=-1)
# Convert channel order if needed
# If the stored order doesn't match the decoder order, flip channels
if img.shape[-1] == 3 and self.channel_order != decoder_order:
img = img[..., ::-1] # Flip RGB <-> BGR
return img
def has_frame(self, frame_idx: int) -> bool:
"""Check if a frame index is contained in the video.
Args:
frame_idx: Index of frame to check.
Returns:
`True` if the index is contained in the video, otherwise `False`.
"""
if self.frame_map:
return frame_idx in self.frame_map
else:
return frame_idx < len(self)
def get_frame_raw_bytes(self, frame_idx: int) -> np.ndarray | None:
"""Get raw encoded bytes for a frame without decoding.
This method reads the raw compressed image data (PNG/JPEG bytes) directly
from the HDF5 dataset without decoding it. This is useful for fast copying
of embedded images when the target format matches the source format.
Args:
frame_idx: Index of the frame to read.
Returns:
Raw encoded bytes as int8 numpy array, or None if:
- The backend doesn't have embedded images (including "hdf5" format which
stores raw numpy arrays, not encoded images)
- The frame index is not available
Notes:
For variable-length datasets, returns the raw bytes directly.
For fixed-length datasets, returns bytes with trailing zeros stripped.
"""
if not self.has_embedded_images:
return None
if not self.has_frame(frame_idx):
return None
# Get the internal index (handle frame_map)
internal_idx = (
self.frame_map.get(frame_idx, frame_idx) if self.frame_map else frame_idx
)
# Read directly from dataset
if self.keep_open:
if self._open_reader is None:
self._open_reader = self._open_h5()
f = self._open_reader
else:
f = self._open_h5()
ds = f[self.dataset]
raw_bytes = ds[internal_idx]
# Handle fixed-length padding (strip trailing zeros)
is_vlen = h5py.check_vlen_dtype(ds.dtype) is not None
if not is_vlen:
# Find last non-zero byte
non_zero_mask = raw_bytes != 0
if non_zero_mask.any():
last_non_zero = np.where(non_zero_mask)[0][-1]
raw_bytes = raw_bytes[: last_non_zero + 1]
if not self.keep_open:
f.close()
return raw_bytes
def _read_frame(self, frame_idx: int) -> np.ndarray:
"""Read a single frame from the video.
Args:
frame_idx: Index of frame to read.
Returns:
The frame as a numpy array of shape `(height, width, channels)`.
Notes:
This does not apply grayscale conversion. It is recommended to use the
`get_frame` method of the `VideoBackend` class instead.
"""
if self.keep_open:
if self._open_reader is None:
self._open_reader = self._open_h5()
f = self._open_reader
else:
f = self._open_h5()
ds = f[self.dataset]
if self.frame_map:
frame_idx = self.frame_map[frame_idx]
img = ds[frame_idx]
if self.has_embedded_images:
img = self.decode_embedded(img)
if self.input_format == "channels_first":
img = np.transpose(img, (2, 1, 0))
if not self.keep_open:
f.close()
return img
def _read_frames(self, frame_inds: list) -> np.ndarray:
"""Read a list of frames from the video.
Args:
frame_inds: List of indices of frames to read.
Returns:
The frame as a numpy array of shape `(frames, height, width, channels)`.
Notes:
This does not apply grayscale conversion. It is recommended to use the
`get_frames` method of the `VideoBackend` class instead.
"""
if self.keep_open:
if self._open_reader is None:
self._open_reader = self._open_h5()
f = self._open_reader
else:
f = self._open_h5()
if self.frame_map:
frame_inds = [self.frame_map[idx] for idx in frame_inds]
ds = f[self.dataset]
imgs = ds[frame_inds]
if "format" in ds.attrs:
imgs = np.stack(
[self.decode_embedded(img) for img in imgs],
axis=0,
)
if self.input_format == "channels_first":
imgs = np.transpose(imgs, (0, 3, 2, 1))
if not self.keep_open:
f.close()
return imgs
@property
def _can_push_crop(self) -> bool:
"""Whether this dataset supports HDF5 crop pushdown (dataset-level gate).
Pushdown reads only a spatial hyperslab of a frame instead of decoding the
whole frame, but it is only valid (and beneficial) for raw rank-4 chunked
datasets with sub-frame spatial chunking and no embedded/frame-mapped
subset. This is the dataset-level gate only (the per-call "crop smaller than
the chunk span" predicate is evaluated in :meth:`read_crop`/:meth:`read_crops`).
The probe reflects the immutable on-disk layout, so the result is cached
after the first call (no file open per read).
Returns:
``True`` if the dataset is a raw (``image_format == "hdf5"``) rank-4
chunked array with sub-frame spatial chunking and an empty
``frame_map``; ``False`` otherwise (including any error while probing,
so a non-applicable dataset never raises).
"""
# Cheap short-circuit (no file open) for embedded/frame-mapped datasets.
if self.image_format != "hdf5" or self.frame_map:
return False
if self._can_push_crop_cached is None:
self._can_push_crop_cached = self._probe_can_push_crop()
return self._can_push_crop_cached
def _probe_can_push_crop(self) -> bool:
"""Probe the on-disk layout for pushdown eligibility (opens the file once)."""
try:
with self._open_h5() as f:
ds = f[self.dataset]
if ds.ndim != 4 or ds.chunks is None:
return False
chunks = ds.chunks
if self.input_format == "channels_first":
# On-disk layout is (F, C, W, H).
disk_w, disk_h = ds.shape[2], ds.shape[3]
return chunks[2] < disk_w or chunks[3] < disk_h
# channels_last on-disk layout is (F, H, W, C).
height, width = ds.shape[1], ds.shape[2]
return chunks[1] < height or chunks[2] < width
except (OSError, KeyError, TypeError): # pragma: no cover - defensive
return False
def read_crop(
self,
frame_idx: int,
crop: tuple[int, int, int, int],
fill: int | tuple[int, ...] = 0,
) -> np.ndarray | None:
"""Read a spatial hyperslab of a frame, padded to the crop shape.
This is the single-frame HDF5 crop pushdown hook consumed by
:class:`CropVideoBackend`. When applicable, it reads only the spatial
region of the frame that overlaps ``crop`` directly from the chunked
dataset (avoiding a full-frame decode) and pads out-of-bounds regions
exactly as :func:`sleap_io.transform.frame.crop_frame` would.
Args:
frame_idx: Index of the frame to read (source-video index; mapped
through ``frame_map`` if present, though pushdown is gated off when
a ``frame_map`` exists).
crop: Crop region ``(x1, y1, x2, y2)`` with ``x2``/``y2`` exclusive.
May be negative or exceed the frame bounds (padded with ``fill``).
fill: Fill value for out-of-bounds regions.
Returns:
A ``(y2 - y1, x2 - x1, C)`` array (pre-grayscale, ``dtype == ds.dtype``)
byte-identical to ``crop_frame(self._read_frame(frame_idx), crop,
fill)`` when pushdown is applicable; otherwise ``None`` to signal the
caller should fall back to a full-frame decode plus ``crop_frame``.
Never raises for out-of-bounds crops.
"""
if not self._can_push_crop:
return None
try:
if self.keep_open:
if self._open_reader is None:
self._open_reader = self._open_h5()
f = self._open_reader
ds = f[self.dataset]
return self._read_crop_from_ds(ds, frame_idx, crop, fill)
else:
with self._open_h5() as f:
ds = f[self.dataset]
return self._read_crop_from_ds(ds, frame_idx, crop, fill)
except (OSError, KeyError, IndexError): # pragma: no cover - defensive
return None
def read_crops(
self,
frame_inds: list,
crop: tuple[int, int, int, int],
fill: int | tuple[int, ...] = 0,
) -> np.ndarray | None:
"""Batched :meth:`read_crop`.
Args:
frame_inds: List of source-video frame indices to read.
crop: Crop region ``(x1, y1, x2, y2)`` with ``x2``/``y2`` exclusive.
fill: Fill value for out-of-bounds regions.
Returns:
A ``(N, y2 - y1, x2 - x1, C)`` array byte-identical to stacking
per-frame ``crop_frame`` results, or ``None`` to fall back to a
full-frame decode plus ``crop_frame``.
"""
if not self._can_push_crop:
return None
try:
if self.keep_open:
if self._open_reader is None:
self._open_reader = self._open_h5()
f = self._open_reader
ds = f[self.dataset]
return self._stack_crops(ds, frame_inds, crop, fill)
else:
with self._open_h5() as f:
ds = f[self.dataset]
return self._stack_crops(ds, frame_inds, crop, fill)
except (OSError, KeyError, IndexError): # pragma: no cover - defensive
return None
def _stack_crops(
self,
ds: h5py.Dataset,
frame_inds: list,
crop: tuple[int, int, int, int],
fill: int | tuple[int, ...],
) -> np.ndarray | None:
"""Stack per-frame crop reads, falling back to ``None`` if the gate declines.
The per-call gate in :meth:`_read_crop_from_ds` is frame-index independent, so
a batch is uniformly all-arrays or all-``None``; returning ``None`` on any
``None`` keeps batched reads byte-for-byte consistent with the scalar path
(the caller then decodes the full frames and crops them).
Args:
ds: The open ``h5py.Dataset`` (raw rank-4).
frame_inds: Frame indices to read.
crop: Crop region ``(x1, y1, x2, y2)``.
fill: Fill value for out-of-bounds regions.
Returns:
A ``(N, y2 - y1, x2 - x1, C)`` array, or ``None`` to signal fallback.
"""
parts = [self._read_crop_from_ds(ds, i, crop, fill) for i in frame_inds]
if any(p is None for p in parts):
return None
return np.stack(parts, axis=0)
def _read_crop_from_ds(
self,
ds: h5py.Dataset,
frame_idx: int,
crop: tuple[int, int, int, int],
fill: int | tuple[int, ...],
) -> np.ndarray | None:
"""Read and pad one frame's crop region from an open dataset.
Performs the per-call gate (crop must be smaller than the chunk span on at
least one spatial axis) and the clamp+pad hyperslab read. Axis ordering is
derived from ``ds.shape`` rather than assumed. Pushdown is structurally gated
off for frame-mapped/embedded datasets (see :attr:`_can_push_crop`), so
``frame_idx`` is always a raw source index here.
Args:
ds: The open ``h5py.Dataset`` (raw rank-4).
frame_idx: Frame index (raw source index; no ``frame_map`` remap needed).
crop: Crop region ``(x1, y1, x2, y2)``.
fill: Fill value for out-of-bounds regions.
Returns:
The ``(y2 - y1, x2 - x1, C)`` cropped/padded frame, or ``None`` if the
per-call gate decides a full read is at least as good.
"""
x1, y1, x2, y2 = crop
chunks = ds.chunks
if self.input_format == "channels_first":
# On-disk layout (F, C, W, H): x maps to axis 2 (W), y to axis 3 (H).
channels = ds.shape[1]
disk_w, disk_h = ds.shape[2], ds.shape[3]
width, height = disk_w, disk_h
chunk_w, chunk_h = chunks[2], chunks[3]
else:
# channels_last (F, H, W, C).
height, width, channels = ds.shape[1], ds.shape[2], ds.shape[3]
chunk_h, chunk_w = chunks[1], chunks[2]
# Per-call gate: only push down when the crop touches fewer spatial chunks
# than the full frame does on at least one axis. If the (in-bounds) crop
# already touches every chunk on both spatial axes, a hyperslab read buys
# nothing over a full read, so fall back.
crop_w, crop_h = x2 - x1, y2 - y1
in_sx1, in_sy1 = max(0, x1), max(0, y1)
in_sx2, in_sy2 = min(width, x2), min(height, y2)
if in_sx2 <= in_sx1 or in_sy2 <= in_sy1:
# Fully outside on at least one axis: no valid source pixels to read,
# so the hyperslab touches no chunks; pushdown is trivially beneficial.
n_chunks_w = n_chunks_h = 0
else:
n_chunks_w = (in_sx2 - 1) // chunk_w - in_sx1 // chunk_w + 1
n_chunks_h = (in_sy2 - 1) // chunk_h - in_sy1 // chunk_h + 1
frame_chunks_w = -(-width // chunk_w)
frame_chunks_h = -(-height // chunk_h)
if n_chunks_w >= frame_chunks_w and n_chunks_h >= frame_chunks_h:
return None
# frame_map is always empty here: _can_push_crop gates pushdown off for
# frame-mapped/embedded datasets, so frame_idx is a raw source index.
out = np.full((crop_h, crop_w, channels), fill, dtype=ds.dtype)
# Clamp the requested rect to the valid frame bounds.
sx1, sy1 = max(0, x1), max(0, y1)
sx2, sy2 = min(width, x2), min(height, y2)
if sx2 > sx1 and sy2 > sy1:
if self.input_format == "channels_first":
region = np.transpose(ds[frame_idx, :, sx1:sx2, sy1:sy2], (2, 1, 0))
else:
region = ds[frame_idx, sy1:sy2, sx1:sx2, :]
out[
sy1 - y1 : sy1 - y1 + (sy2 - sy1),
sx1 - x1 : sx1 - x1 + (sx2 - sx1),
] = region
return out
EXTS = ('h5', 'hdf5', 'slp')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__annotations__ = {'dataset': 'str | None', 'input_format': 'str', 'frame_map': 'dict[int, int]', '_can_push_crop_cached': 'bool | None', 'source_filename': 'str | None', 'source_inds': 'np.ndarray | None', 'image_format': 'str', 'channel_order': 'str', 'plugin': 'str | None', '_url_file': 'object | None', '_url_headers': 'dict[str, str] | None', '_url_stream_mode': 'str'}
class-attribute
¶
dict() -> new empty dictionary dict(mapping) -> new dictionary initialized from a mapping object's (key, value) pairs dict(iterable) -> new dictionary initialized as if via: d = {} for k, v in iterable: d[k] = v dict(**kwargs) -> new dictionary initialized with the name=value pairs in the keyword argument list. For example: dict(one=1, two=2)
__attrs_own_setattr__ = True
class-attribute
¶
Returns True when the argument is true, False otherwise. The builtins True and False are the only two instances of the class bool. The class bool is a subclass of the class int, and cannot be subclassed.
__attrs_props__ = ClassProps(is_exception=False, is_slotted=True, has_weakref_slot=True, is_frozen=False, kw_only=<KeywordOnly.NO: 'no'>, collected_fields_by_mro=True, added_init=True, added_repr=True, added_eq=True, added_ordering=False, hashability=<Hashability.UNHASHABLE: 'unhashable'>, added_match_args=True, added_str=False, added_pickling=False, on_setattr_hook=<function pipe.<locals>.wrapped_pipe at 0x7f08471ce840>, field_transformer=None)
class-attribute
¶
Effective class properties as derived from parameters to attr.s() or
define() decorators.
This is the same data structure that attrs uses internally to decide how to construct the final class.
Warning:
This feature is currently **experimental** and is not covered by our
strict backwards-compatibility guarantees.
Attributes:
| Name | Type | Description |
|---|---|---|
is_exception |
bool
|
Whether the class is treated as an exception class. |
is_slotted |
bool
|
Whether the class is |
has_weakref_slot |
bool
|
Whether the class has a slot for weak references. |
is_frozen |
bool
|
Whether the class is frozen. |
kw_only |
KeywordOnly
|
Whether / how the class enforces keyword-only arguments on the
|
collected_fields_by_mro |
bool
|
Whether the class fields were collected by method resolution order.
That is, correctly but unlike |
added_init |
bool
|
Whether the class has an attrs-generated |
added_repr |
bool
|
Whether the class has an attrs-generated |
added_eq |
bool
|
Whether the class has attrs-generated equality methods. |
added_ordering |
bool
|
Whether the class has attrs-generated ordering methods. |
hashability |
Hashability
|
How |
added_match_args |
bool
|
Whether the class supports positional |
added_str |
bool
|
Whether the class has an attrs-generated |
added_pickling |
bool
|
Whether the class has attrs-generated |
on_setattr_hook |
Callable[[Any, Attribute[Any], Any], Any] | None
|
The class's |
field_transformer |
Callable[[Attribute[Any]], Attribute[Any]] | None
|
The class's |
.. versionadded:: 25.4.0
__doc__ = 'Video backend for reading videos stored in HDF5 files.\n\nThis backend supports reading videos stored in HDF5 files, both in rank-4 datasets\nas well as in datasets with lists of binary-encoded images.\n\nEmbedded image datasets are used in SLEAP when exporting package files (`.pkg.slp`)\nwith videos embedded in them. This is useful for bundling training or inference data\nwithout having to worry about the videos (or frame images) being moved or deleted.\nIt is expected that these types of datasets will be in a `Group` with a `int8`\nvariable length dataset called `"video"`. This dataset must also contain an\nattribute called "format" with a string describing the image format (e.g., "png" or\n"jpg") which will be used to decode it appropriately.\n\nIf a `frame_numbers` dataset is present in the group, it will be used to map from\nsource video frames to the frames in the dataset. This is useful to preserve frame\nindexing when exporting a subset of frames in the video. It will also be used to\npopulate `frame_map` and `source_inds` attributes.\n\nAttributes:\n filename: Path to HDF5 file (.h5, .hdf5 or .slp).\n grayscale: Whether to force grayscale. If None, autodetect on first frame load.\n keep_open: Whether to keep the video reader open between calls to read frames.\n If False, will close the reader after each call. If True (the default), it\n will keep the reader open and cache it for subsequent calls which may\n enhance the performance of reading multiple frames.\n dataset: Name of dataset to read from. If `None`, will try to find a rank-4\n dataset by iterating through datasets in the file. If specifying an embedded\n dataset, this can be the group containing a "video" dataset or the dataset\n itself (e.g., "video0" or "video0/video").\n input_format: Format of the data in the dataset. One of "channels_last" (the\n default) in `(frames, height, width, channels)` order or "channels_first" in\n `(frames, channels, width, height)` order. Embedded datasets should use the\n "channels_last" format.\n frame_map: Mapping from frame indices to indices in the dataset. This is used to\n translate between the frame indices of the images within their source video\n and the indices of the images in the dataset. This is only used when reading\n embedded image datasets.\n source_filename: Path to the source video file. This is metadata and only used\n when reading embedded image datasets.\n source_inds: Indices of the frames in the source video file. This is metadata\n and only used when reading embedded image datasets.\n image_format: Format of the images in the embedded dataset. This is metadata and\n only used when reading embedded image datasets.\n channel_order: Channel order of embedded images, either "RGB" or "BGR". This is\n used to ensure consistent color channel ordering when decoding embedded\n images. If the encoding and decoding plugins have different channel orders,\n the channels will be automatically flipped during decoding.\n plugin: Plugin to use for decoding embedded images. One of "opencv" or\n "FFMPEG". If None, uses the global default or auto-detects based on\n available packages. Note that "pyav" is automatically mapped to "FFMPEG"\n since PyAV doesn\'t support image decoding.\n\nNotes:\n Concurrent reads of a single remote (URL-backed) `HDF5Video` from\n multiple threads are safe: although all reads share one cached fsspec\n file-like (a single byte position), h5py serializes every HDF5 C-library\n call under a global recursive lock (`h5py._objects.phil`), so the\n seek+read pair a frame read performs is never interleaved across threads.\n For true read *parallelism* (rather than just safety), construct\n independent `Video`/`HDF5Video` instances per worker; each gets its own\n fsspec file and block cache.\n'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__firstlineno__ = 1126
class-attribute
¶
int([x]) -> integer int(x, base=10) -> integer
Convert a number or string to an integer, or return 0 if no arguments are given. If x is a number, return x.int(). For floating-point numbers, this truncates towards zero.
If x is not a number or if base is given, then x must be a string, bytes, or bytearray instance representing an integer literal in the given base. The literal can be preceded by '+' or '-' and be surrounded by whitespace. The base defaults to 10. Valid bases are 0 and 2-36. Base 0 means to interpret the base from the string as an integer iteral.
int('0b100', base=0) 4
__match_args__ = ('filename', 'grayscale', 'keep_open', '_cached_shape', '_open_reader', '_fps', 'dataset', 'input_format', 'source_filename', 'source_inds', 'image_format', 'channel_order', 'plugin', '_url_headers', '_url_stream_mode')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__module__ = 'sleap_io.io.video_reading'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__slots__ = ('dataset', 'input_format', 'frame_map', '_can_push_crop_cached', 'source_filename', 'source_inds', 'image_format', 'channel_order', 'plugin', '_url_file', '_url_headers', '_url_stream_mode')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__static_attributes__ = ('_can_push_crop_cached', '_fps', '_open_reader', '_url_file', 'channel_order', 'dataset', 'frame_map', 'image_format', 'plugin', 'source_filename', 'source_inds')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
embedded_frame_inds
property
¶
Return the frame indices of the embedded images.
has_embedded_images
property
¶
Return True if the dataset contains embedded images.
img_shape
property
¶
Shape of a single frame in the video as (height, width, channels).
num_frames
property
¶
Number of frames in the video.
__attrs_post_init__()
¶
Auto-detect dataset and frame map heuristically.
Source code in sleap_io/io/video_reading.py
def __attrs_post_init__(self):
"""Auto-detect dataset and frame map heuristically."""
# Check if the file accessible before applying heuristics.
# For URLs, track whether this probe opened the cached fsspec file-like
# so it can be released afterwards (it would otherwise leak the handle on
# an early return / exception, and a probe-time open may predate the
# final auth headers being applied).
url_file_preexisting = self._url_file is not None
try:
f = self._open_h5()
except OSError:
self._release_probe_url_file(url_file_preexisting)
return
try:
if self.dataset is None:
# Iterate through datasets to find a rank 4 array.
def find_movies(name, obj):
if isinstance(obj, h5py.Dataset) and obj.ndim == 4:
self.dataset = name
return True
f.visititems(find_movies)
if self.dataset is None:
# Iterate through datasets to find an embedded video dataset.
def find_embedded(name, obj):
if isinstance(obj, h5py.Dataset) and name.endswith("/video"):
self.dataset = name
return True
f.visititems(find_embedded)
if self.dataset is None:
# Couldn't find video datasets.
return
if isinstance(f[self.dataset], h5py.Group):
# If this is a group, assume it's an embedded video dataset.
if "video" in f[self.dataset]:
self.dataset = f"{self.dataset}/video"
if self.dataset.split("/")[-1] == "video":
# This may be an embedded video dataset. Check for frame map.
ds = f[self.dataset]
if "format" in ds.attrs:
self.image_format = ds.attrs["format"]
# Read channel_order, with backwards compatibility
if "channel_order" in ds.attrs:
self.channel_order = ds.attrs["channel_order"]
else:
# Backwards compatibility: Check format_id for older files
# Prior to format 1.4, embedded images were primarily encoded
# with OpenCV which uses BGR, so default to BGR for older
# formats
if "metadata" in f and "format_id" in f["metadata"].attrs:
format_id = f["metadata"].attrs["format_id"]
if format_id < 1.4:
self.channel_order = "BGR" # Legacy default
# If no format_id found, assume BGR (safest legacy default)
# since most embedded images before this change used OpenCV
if "frame_numbers" in ds.parent:
frame_numbers = ds.parent["frame_numbers"][:].astype(int)
self.frame_map = {
frame: idx for idx, frame in enumerate(frame_numbers)
}
self.source_inds = frame_numbers
if "source_video" in ds.parent:
source_grp = ds.parent["source_video"]
# Source metadata is normally in the "json" attribute, but
# oversized metadata (e.g. an image-sequence source with many
# thousands of filenames, exceeding HDF5's 64 KB attribute limit)
# is stored in a "json" *dataset* instead (see
# ``slp._write_source_video_json``). Read whichever is present so
# such packages remain openable -- otherwise the backend fails to
# open, ``Video.backend`` is left ``None``, and embedded frames
# cannot be read.
if "json" in source_grp:
source_json = source_grp["json"][()]
else:
source_json = source_grp.attrs["json"]
self.source_filename = json.loads(source_json)["backend"][
"filename"
]
# Read FPS from attributes if present
if "fps" in ds.attrs:
self._fps = float(ds.attrs["fps"])
elif "fps" in ds.parent.attrs:
self._fps = float(ds.parent.attrs["fps"])
finally:
f.close()
self._release_probe_url_file(url_file_preexisting)
# Set default plugin if not specified (use image plugin, not video plugin)
if self.plugin is None:
# Check image plugin default first (for embedded images)
if _default_image_plugin is not None:
self.plugin = _default_image_plugin
# Otherwise auto-detect (for embedded image decoding)
elif "cv2" in sys.modules:
self.plugin = "opencv"
else:
self.plugin = "imageio" # imageio fallback
__eq__(other)
¶
Method generated by attrs for class HDF5Video.
Source code in sleap_io/io/video_reading.py
from sleap_io.io import _remote
from sleap_io.transform.frame import crop_frame
from sleap_io.transform.points import crop_points, uncrop_points
try:
import cv2
except ImportError:
pass
try:
import imageio_ffmpeg # noqa: F401
except ImportError:
pass
try:
import av # noqa: F401
except ImportError:
pass
__getstate__()
¶
Return state for pickling/deepcopy, dropping unpicklable handles.
Extends :meth:VideoBackend.__getstate__ to also drop the cached
fsspec-backed _url_file (reopened lazily by :meth:_open_h5).
Source code in sleap_io/io/video_reading.py
def __getstate__(self) -> dict:
"""Return state for pickling/deepcopy, dropping unpicklable handles.
Extends :meth:`VideoBackend.__getstate__` to also drop the cached
fsspec-backed ``_url_file`` (reopened lazily by :meth:`_open_h5`).
"""
state = super().__getstate__()
state["_url_file"] = None
return state
__init__(filename, grayscale=None, keep_open=True, cached_shape=None, open_reader=None, fps=None, dataset=None, input_format='channels_last', source_filename=None, source_inds=None, image_format='hdf5', channel_order='RGB', plugin=None, url_headers=None, url_stream_mode='blockcache')
¶
Method generated by attrs for class HDF5Video.
Source code in sleap_io/io/video_reading.py
# Track available backends (populated on module import)
_AVAILABLE_VIDEO_BACKENDS = {
"opencv": "cv2" in sys.modules,
"FFMPEG": "imageio_ffmpeg" in sys.modules,
"pyav": "av" in sys.modules,
}
_AVAILABLE_IMAGE_BACKENDS = {
"opencv": "cv2" in sys.modules,
"imageio": True, # Always available (core dependency)
}
# Global default video plugin
_default_video_plugin: str | None = None
def normalize_plugin_name(plugin: str) -> str:
"""Normalize plugin names to standard format.
Args:
plugin: Plugin name or alias (case-insensitive).
__repr__()
¶
Method generated by attrs for class HDF5Video.
__setattr__(name, val)
¶
Method generated by attrs for class HDF5Video.
Source code in sleap_io/io/video_reading.py
multiple threads are safe: although all reads share one cached fsspec
file-like (a single byte position), h5py serializes every HDF5 C-library
call under a global recursive lock (`h5py._objects.phil`), so the
seek+read pair a frame read performs is never interleaved across threads.
For true read *parallelism* (rather than just safety), construct
independent `Video`/`HDF5Video` instances per worker; each gets its own
fsspec file and block cache.
"""
close()
¶
Release the cached HDF5 reader and the cached fsspec URL file-like.
Extends :meth:VideoBackend.close (which drops the cached h5py.File
reader) by also closing the fsspec-backed _url_file shared across
reads, which h5py.File.close() does not close on its own. Both are
lazily reopened on the next read, so this is safe to call between reads.
Source code in sleap_io/io/video_reading.py
def close(self) -> None:
"""Release the cached HDF5 reader and the cached fsspec URL file-like.
Extends :meth:`VideoBackend.close` (which drops the cached ``h5py.File``
reader) by also closing the fsspec-backed ``_url_file`` shared across
reads, which ``h5py.File.close()`` does not close on its own. Both are
lazily reopened on the next read, so this is safe to call between reads.
"""
super().close()
self._close_url_file()
decode_embedded(img_string)
¶
Decode an embedded image string into a numpy array.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
img_string
|
ndarray
|
Binary string of the image as a |
required |
Returns:
| Type | Description |
|---|---|
ndarray
|
The decoded image as a numpy array of shape This method does not apply grayscale conversion as per the |
Source code in sleap_io/io/video_reading.py
def decode_embedded(self, img_string: np.ndarray) -> np.ndarray:
"""Decode an embedded image string into a numpy array.
Args:
img_string: Binary string of the image as a `int8` numpy vector with the
bytes as values corresponding to the format-encoded image.
Returns:
The decoded image as a numpy array of shape `(height, width, channels)`. If
a rank-2 image is decoded, it will be expanded such that channels will be 1.
This method does not apply grayscale conversion as per the `grayscale`
attribute. Use the `get_frame` or `get_frames` methods of the `VideoBackend`
to apply grayscale conversion rather than calling this function directly.
"""
# Decode based on plugin
if self.plugin == "opencv":
img = cv2.imdecode(img_string, cv2.IMREAD_UNCHANGED)
decoder_order = "BGR" # OpenCV decodes to BGR
else:
# Use imageio for FFMPEG or any other plugin
img = iio.imread(BytesIO(img_string), extension=f".{self.image_format}")
decoder_order = "RGB" # imageio decodes to RGB
if img.ndim == 2:
img = np.expand_dims(img, axis=-1)
# Convert channel order if needed
# If the stored order doesn't match the decoder order, flip channels
if img.shape[-1] == 3 and self.channel_order != decoder_order:
img = img[..., ::-1] # Flip RGB <-> BGR
return img
get_frame_raw_bytes(frame_idx)
¶
Get raw encoded bytes for a frame without decoding.
This method reads the raw compressed image data (PNG/JPEG bytes) directly from the HDF5 dataset without decoding it. This is useful for fast copying of embedded images when the target format matches the source format.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame_idx
|
int
|
Index of the frame to read. |
required |
Returns:
| Type | Description |
|---|---|
ndarray | None
|
Raw encoded bytes as int8 numpy array, or None if: - The backend doesn't have embedded images (including "hdf5" format which stores raw numpy arrays, not encoded images) - The frame index is not available |
Notes
For variable-length datasets, returns the raw bytes directly. For fixed-length datasets, returns bytes with trailing zeros stripped.
Source code in sleap_io/io/video_reading.py
def get_frame_raw_bytes(self, frame_idx: int) -> np.ndarray | None:
"""Get raw encoded bytes for a frame without decoding.
This method reads the raw compressed image data (PNG/JPEG bytes) directly
from the HDF5 dataset without decoding it. This is useful for fast copying
of embedded images when the target format matches the source format.
Args:
frame_idx: Index of the frame to read.
Returns:
Raw encoded bytes as int8 numpy array, or None if:
- The backend doesn't have embedded images (including "hdf5" format which
stores raw numpy arrays, not encoded images)
- The frame index is not available
Notes:
For variable-length datasets, returns the raw bytes directly.
For fixed-length datasets, returns bytes with trailing zeros stripped.
"""
if not self.has_embedded_images:
return None
if not self.has_frame(frame_idx):
return None
# Get the internal index (handle frame_map)
internal_idx = (
self.frame_map.get(frame_idx, frame_idx) if self.frame_map else frame_idx
)
# Read directly from dataset
if self.keep_open:
if self._open_reader is None:
self._open_reader = self._open_h5()
f = self._open_reader
else:
f = self._open_h5()
ds = f[self.dataset]
raw_bytes = ds[internal_idx]
# Handle fixed-length padding (strip trailing zeros)
is_vlen = h5py.check_vlen_dtype(ds.dtype) is not None
if not is_vlen:
# Find last non-zero byte
non_zero_mask = raw_bytes != 0
if non_zero_mask.any():
last_non_zero = np.where(non_zero_mask)[0][-1]
raw_bytes = raw_bytes[: last_non_zero + 1]
if not self.keep_open:
f.close()
return raw_bytes
has_frame(frame_idx)
¶
Check if a frame index is contained in the video.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame_idx
|
int
|
Index of frame to check. |
required |
Returns:
| Type | Description |
|---|---|
bool
|
|
Source code in sleap_io/io/video_reading.py
def has_frame(self, frame_idx: int) -> bool:
"""Check if a frame index is contained in the video.
Args:
frame_idx: Index of frame to check.
Returns:
`True` if the index is contained in the video, otherwise `False`.
"""
if self.frame_map:
return frame_idx in self.frame_map
else:
return frame_idx < len(self)
read_crop(frame_idx, crop, fill=0)
¶
Read a spatial hyperslab of a frame, padded to the crop shape.
This is the single-frame HDF5 crop pushdown hook consumed by
:class:CropVideoBackend. When applicable, it reads only the spatial
region of the frame that overlaps crop directly from the chunked
dataset (avoiding a full-frame decode) and pads out-of-bounds regions
exactly as :func:sleap_io.transform.frame.crop_frame would.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame_idx
|
int
|
Index of the frame to read (source-video index; mapped
through |
required |
crop
|
tuple[int, int, int, int]
|
Crop region |
required |
fill
|
int | tuple[int, ...]
|
Fill value for out-of-bounds regions. |
0
|
Returns:
| Type | Description |
|---|---|
ndarray | None
|
A |
Source code in sleap_io/io/video_reading.py
def read_crop(
self,
frame_idx: int,
crop: tuple[int, int, int, int],
fill: int | tuple[int, ...] = 0,
) -> np.ndarray | None:
"""Read a spatial hyperslab of a frame, padded to the crop shape.
This is the single-frame HDF5 crop pushdown hook consumed by
:class:`CropVideoBackend`. When applicable, it reads only the spatial
region of the frame that overlaps ``crop`` directly from the chunked
dataset (avoiding a full-frame decode) and pads out-of-bounds regions
exactly as :func:`sleap_io.transform.frame.crop_frame` would.
Args:
frame_idx: Index of the frame to read (source-video index; mapped
through ``frame_map`` if present, though pushdown is gated off when
a ``frame_map`` exists).
crop: Crop region ``(x1, y1, x2, y2)`` with ``x2``/``y2`` exclusive.
May be negative or exceed the frame bounds (padded with ``fill``).
fill: Fill value for out-of-bounds regions.
Returns:
A ``(y2 - y1, x2 - x1, C)`` array (pre-grayscale, ``dtype == ds.dtype``)
byte-identical to ``crop_frame(self._read_frame(frame_idx), crop,
fill)`` when pushdown is applicable; otherwise ``None`` to signal the
caller should fall back to a full-frame decode plus ``crop_frame``.
Never raises for out-of-bounds crops.
"""
if not self._can_push_crop:
return None
try:
if self.keep_open:
if self._open_reader is None:
self._open_reader = self._open_h5()
f = self._open_reader
ds = f[self.dataset]
return self._read_crop_from_ds(ds, frame_idx, crop, fill)
else:
with self._open_h5() as f:
ds = f[self.dataset]
return self._read_crop_from_ds(ds, frame_idx, crop, fill)
except (OSError, KeyError, IndexError): # pragma: no cover - defensive
return None
read_crops(frame_inds, crop, fill=0)
¶
Batched :meth:read_crop.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame_inds
|
list
|
List of source-video frame indices to read. |
required |
crop
|
tuple[int, int, int, int]
|
Crop region |
required |
fill
|
int | tuple[int, ...]
|
Fill value for out-of-bounds regions. |
0
|
Returns:
| Type | Description |
|---|---|
ndarray | None
|
A |
Source code in sleap_io/io/video_reading.py
def read_crops(
self,
frame_inds: list,
crop: tuple[int, int, int, int],
fill: int | tuple[int, ...] = 0,
) -> np.ndarray | None:
"""Batched :meth:`read_crop`.
Args:
frame_inds: List of source-video frame indices to read.
crop: Crop region ``(x1, y1, x2, y2)`` with ``x2``/``y2`` exclusive.
fill: Fill value for out-of-bounds regions.
Returns:
A ``(N, y2 - y1, x2 - x1, C)`` array byte-identical to stacking
per-frame ``crop_frame`` results, or ``None`` to fall back to a
full-frame decode plus ``crop_frame``.
"""
if not self._can_push_crop:
return None
try:
if self.keep_open:
if self._open_reader is None:
self._open_reader = self._open_h5()
f = self._open_reader
ds = f[self.dataset]
return self._stack_crops(ds, frame_inds, crop, fill)
else:
with self._open_h5() as f:
ds = f[self.dataset]
return self._stack_crops(ds, frame_inds, crop, fill)
except (OSError, KeyError, IndexError): # pragma: no cover - defensive
return None
read_test_frame()
¶
Read a single frame from the video to test for grayscale.
ImageVideo
¶
Bases: sleap_io.io.video_reading.VideoBackend
Video backend for reading videos stored as image files.
This backend supports reading videos stored as a list of images.
Attributes:
| Name | Type | Description |
|---|---|---|
filename |
Path to image files. |
|
grayscale |
Whether to force grayscale. If None, autodetect on first frame load. |
|
plugin |
Image plugin to use for reading. One of "opencv" or "imageio". If None, uses global default from get_default_image_plugin(), or auto-detects. |
Methods:
| Name | Description |
|---|---|
__eq__ |
Method generated by attrs for class ImageVideo. |
__init__ |
Method generated by attrs for class ImageVideo. |
__repr__ |
Method generated by attrs for class ImageVideo. |
__setattr__ |
Method generated by attrs for class ImageVideo. |
find_images |
Find images in a folder and return a list of filenames. |
get_frame_raw_bytes |
Return the raw encoded bytes of the source image file for a frame. |
Source code in sleap_io/io/video_reading.py
@attrs.define
class ImageVideo(VideoBackend):
"""Video backend for reading videos stored as image files.
This backend supports reading videos stored as a list of images.
Attributes:
filename: Path to image files.
grayscale: Whether to force grayscale. If None, autodetect on first frame load.
plugin: Image plugin to use for reading. One of "opencv" or "imageio".
If None, uses global default from get_default_image_plugin(), or
auto-detects.
"""
EXTS = ("png", "jpg", "jpeg", "tif", "tiff", "bmp")
plugin: str = attrs.field()
@plugin.validator
def _validate_plugin(self, attribute, value):
"""Validate and normalize plugin name."""
normalized = normalize_image_plugin_name(value)
object.__setattr__(self, attribute.name, normalized)
@plugin.default
def _default_plugin(self) -> str:
"""Get default plugin, checking global default first."""
# Check global default first
if _default_image_plugin is not None:
# Warn if preferred plugin not available
if not _AVAILABLE_IMAGE_BACKENDS.get(_default_image_plugin, False):
import warnings
available = get_available_image_backends()
install_cmd = get_installation_instructions(
_default_image_plugin, "image"
)
warnings.warn(
f"Preferred image plugin '{_default_image_plugin}' is not "
f"available. Available plugins: {available}\n"
f"Install with: {install_cmd}"
)
# Fall through to auto-detection
else:
return _default_image_plugin
# Otherwise auto-detect
if "cv2" in sys.modules:
return "opencv"
else:
return "imageio"
@staticmethod
def find_images(folder: str) -> list[str]:
"""Find images in a folder and return a list of filenames."""
folder = Path(folder)
return sorted(
[f.as_posix() for f in folder.glob("*") if f.suffix[1:] in ImageVideo.EXTS]
)
@property
def num_frames(self) -> int:
"""Number of frames in the video."""
return len(self.filename)
def get_frame_raw_bytes(self, frame_idx: int) -> np.ndarray | None:
"""Return the raw encoded bytes of the source image file for a frame.
Reads the on-disk image file verbatim (no decode/re-encode), enabling a
direct byte-for-byte embed of already-compressed sources. This avoids both
the cost of a decode/re-encode cycle and any additional compression
artifacts (important for lossy JPEG sources).
Only PNG/JPEG sources are supported here -- these are already entropy-coded
and are decodable by the embedded-image reader (`HDF5Video.decode_embedded`).
Other extensions (e.g. TIFF/BMP) return `None` so the caller falls back to
decoding and re-encoding to the requested format.
Args:
frame_idx: Index of the frame to read.
Returns:
The raw file bytes as an `int8` numpy vector, or `None` if the source
file is not a directly-storable compressed image (PNG/JPEG) or cannot
be read.
Notes:
Bytes copied this way decode back to RGB (matching `_read_frame`), so
the embedded dataset should record `channel_order="RGB"`.
"""
filename = self.filename[frame_idx]
ext = Path(filename).suffix.lower().lstrip(".")
if ext not in ("png", "jpg", "jpeg"):
return None
try:
with open(filename, "rb") as f:
data = f.read()
except OSError:
return None
return np.frombuffer(data, dtype="int8")
def _read_frame(self, frame_idx: int) -> np.ndarray:
"""Read a single frame from the video.
Args:
frame_idx: Index of frame to read.
Returns:
The frame as a numpy array of shape `(height, width, channels)` in RGB
order.
Notes:
This does not apply grayscale conversion. It is recommended to use the
`get_frame` method of the `VideoBackend` class instead.
Images are always returned in RGB order regardless of plugin:
- imageio: Returns RGB natively
- opencv: Returns BGR, automatically flipped to RGB
"""
if self.plugin == "opencv":
# OpenCV reads as BGR, flip to RGB
img = cv2.imread(self.filename[frame_idx], cv2.IMREAD_UNCHANGED)
if img is None:
raise ValueError(f"Failed to read image: {self.filename[frame_idx]}")
if img.ndim == 3 and img.shape[-1] == 3:
img = img[..., ::-1] # BGR -> RGB
else: # imageio
# imageio reads as RGB natively
img = iio.imread(self.filename[frame_idx])
if img.ndim == 2:
img = np.expand_dims(img, axis=-1)
return img
EXTS = ('png', 'jpg', 'jpeg', 'tif', 'tiff', 'bmp')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__annotations__ = {'plugin': 'str'}
class-attribute
¶
dict() -> new empty dictionary dict(mapping) -> new dictionary initialized from a mapping object's (key, value) pairs dict(iterable) -> new dictionary initialized as if via: d = {} for k, v in iterable: d[k] = v dict(**kwargs) -> new dictionary initialized with the name=value pairs in the keyword argument list. For example: dict(one=1, two=2)
__attrs_own_setattr__ = True
class-attribute
¶
Returns True when the argument is true, False otherwise. The builtins True and False are the only two instances of the class bool. The class bool is a subclass of the class int, and cannot be subclassed.
__attrs_props__ = ClassProps(is_exception=False, is_slotted=True, has_weakref_slot=True, is_frozen=False, kw_only=<KeywordOnly.NO: 'no'>, collected_fields_by_mro=True, added_init=True, added_repr=True, added_eq=True, added_ordering=False, hashability=<Hashability.UNHASHABLE: 'unhashable'>, added_match_args=True, added_str=False, added_pickling=True, on_setattr_hook=<function pipe.<locals>.wrapped_pipe at 0x7f08471ce840>, field_transformer=None)
class-attribute
¶
Effective class properties as derived from parameters to attr.s() or
define() decorators.
This is the same data structure that attrs uses internally to decide how to construct the final class.
Warning:
This feature is currently **experimental** and is not covered by our
strict backwards-compatibility guarantees.
Attributes:
| Name | Type | Description |
|---|---|---|
is_exception |
bool
|
Whether the class is treated as an exception class. |
is_slotted |
bool
|
Whether the class is |
has_weakref_slot |
bool
|
Whether the class has a slot for weak references. |
is_frozen |
bool
|
Whether the class is frozen. |
kw_only |
KeywordOnly
|
Whether / how the class enforces keyword-only arguments on the
|
collected_fields_by_mro |
bool
|
Whether the class fields were collected by method resolution order.
That is, correctly but unlike |
added_init |
bool
|
Whether the class has an attrs-generated |
added_repr |
bool
|
Whether the class has an attrs-generated |
added_eq |
bool
|
Whether the class has attrs-generated equality methods. |
added_ordering |
bool
|
Whether the class has attrs-generated ordering methods. |
hashability |
Hashability
|
How |
added_match_args |
bool
|
Whether the class supports positional |
added_str |
bool
|
Whether the class has an attrs-generated |
added_pickling |
bool
|
Whether the class has attrs-generated |
on_setattr_hook |
Callable[[Any, Attribute[Any], Any], Any] | None
|
The class's |
field_transformer |
Callable[[Attribute[Any]], Attribute[Any]] | None
|
The class's |
.. versionadded:: 25.4.0
__doc__ = 'Video backend for reading videos stored as image files.\n\nThis backend supports reading videos stored as a list of images.\n\nAttributes:\n filename: Path to image files.\n grayscale: Whether to force grayscale. If None, autodetect on first frame load.\n plugin: Image plugin to use for reading. One of "opencv" or "imageio".\n If None, uses global default from get_default_image_plugin(), or\n auto-detects.\n'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__firstlineno__ = 1871
class-attribute
¶
int([x]) -> integer int(x, base=10) -> integer
Convert a number or string to an integer, or return 0 if no arguments are given. If x is a number, return x.int(). For floating-point numbers, this truncates towards zero.
If x is not a number or if base is given, then x must be a string, bytes, or bytearray instance representing an integer literal in the given base. The literal can be preceded by '+' or '-' and be surrounded by whitespace. The base defaults to 10. Valid bases are 0 and 2-36. Base 0 means to interpret the base from the string as an integer iteral.
int('0b100', base=0) 4
__match_args__ = ('filename', 'grayscale', 'keep_open', '_cached_shape', '_open_reader', '_fps', 'plugin')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__module__ = 'sleap_io.io.video_reading'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__slots__ = ('plugin',)
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__static_attributes__ = ()
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
num_frames
property
¶
Number of frames in the video.
__eq__(other)
¶
Method generated by attrs for class ImageVideo.
__init__(filename, grayscale=None, keep_open=True, cached_shape=None, open_reader=None, fps=None, plugin=NOTHING)
¶
Method generated by attrs for class ImageVideo.
__repr__()
¶
Method generated by attrs for class ImageVideo.
__setattr__(name, val)
¶
Method generated by attrs for class ImageVideo.
Source code in sleap_io/io/video_reading.py
multiple threads are safe: although all reads share one cached fsspec
file-like (a single byte position), h5py serializes every HDF5 C-library
call under a global recursive lock (`h5py._objects.phil`), so the
seek+read pair a frame read performs is never interleaved across threads.
For true read *parallelism* (rather than just safety), construct
independent `Video`/`HDF5Video` instances per worker; each gets its own
fsspec file and block cache.
"""
find_images(folder)
staticmethod
¶
Find images in a folder and return a list of filenames.
get_frame_raw_bytes(frame_idx)
¶
Return the raw encoded bytes of the source image file for a frame.
Reads the on-disk image file verbatim (no decode/re-encode), enabling a direct byte-for-byte embed of already-compressed sources. This avoids both the cost of a decode/re-encode cycle and any additional compression artifacts (important for lossy JPEG sources).
Only PNG/JPEG sources are supported here -- these are already entropy-coded
and are decodable by the embedded-image reader (HDF5Video.decode_embedded).
Other extensions (e.g. TIFF/BMP) return None so the caller falls back to
decoding and re-encoding to the requested format.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame_idx
|
int
|
Index of the frame to read. |
required |
Returns:
| Type | Description |
|---|---|
ndarray | None
|
The raw file bytes as an |
Notes
Bytes copied this way decode back to RGB (matching _read_frame), so
the embedded dataset should record channel_order="RGB".
Source code in sleap_io/io/video_reading.py
def get_frame_raw_bytes(self, frame_idx: int) -> np.ndarray | None:
"""Return the raw encoded bytes of the source image file for a frame.
Reads the on-disk image file verbatim (no decode/re-encode), enabling a
direct byte-for-byte embed of already-compressed sources. This avoids both
the cost of a decode/re-encode cycle and any additional compression
artifacts (important for lossy JPEG sources).
Only PNG/JPEG sources are supported here -- these are already entropy-coded
and are decodable by the embedded-image reader (`HDF5Video.decode_embedded`).
Other extensions (e.g. TIFF/BMP) return `None` so the caller falls back to
decoding and re-encoding to the requested format.
Args:
frame_idx: Index of the frame to read.
Returns:
The raw file bytes as an `int8` numpy vector, or `None` if the source
file is not a directly-storable compressed image (PNG/JPEG) or cannot
be read.
Notes:
Bytes copied this way decode back to RGB (matching `_read_frame`), so
the embedded dataset should record `channel_order="RGB"`.
"""
filename = self.filename[frame_idx]
ext = Path(filename).suffix.lower().lstrip(".")
if ext not in ("png", "jpg", "jpeg"):
return None
try:
with open(filename, "rb") as f:
data = f.read()
except OSError:
return None
return np.frombuffer(data, dtype="int8")
MediaVideo
¶
Bases: sleap_io.io.video_reading.VideoBackend
Video backend for reading videos stored as common media files.
This backend supports reading through FFMPEG (the default), pyav, or OpenCV. Here are their trade-offs:
- "opencv": Fastest video reader, but only supports a limited number of codecs
and may not be able to read some videos. It requires `opencv-python` to be
installed. It is the fastest because it uses the OpenCV C++ library to read
videos, but is limited by the version of FFMPEG that was linked into it at
build time as well as the OpenCV version used.
- "FFMPEG": Slowest, but most reliable. This is the default backend. It requires
`imageio-ffmpeg` and a `ffmpeg` executable on the system path (which can be
installed via conda). The `imageio` plugin for FFMPEG reads frames into raw
bytes which are communicated to Python through STDOUT on a subprocess pipe,
which can be slow. However, it is the most reliable and feature-complete. If
you install the conda-forge version of ffmpeg, it will be compiled with
support for many codecs, including GPU-accelerated codecs like NVDEC for
H264 and others.
- "pyav": Supports most codecs that FFMPEG does, but not as complete or reliable
of an implementation in `imageio` as FFMPEG for some video types. It is
faster than FFMPEG because it uses the `av` package to read frames directly
into numpy arrays in memory without the need for a subprocess pipe. These
are Python bindings for the C library libav, which is the same library that
FFMPEG uses under the hood.
Attributes:
| Name | Type | Description |
|---|---|---|
filename |
Path to video file. |
|
grayscale |
Whether to force grayscale. If None, autodetect on first frame load. |
|
keep_open |
Whether to keep the video reader open between calls to read frames. If False, will close the reader after each call. If True (the default), it will keep the reader open and cache it for subsequent calls which may enhance the performance of reading multiple frames. |
|
plugin |
Video plugin to use. One of "opencv", "FFMPEG", or "pyav". If |
Methods:
| Name | Description |
|---|---|
__eq__ |
Method generated by attrs for class MediaVideo. |
__init__ |
Method generated by attrs for class MediaVideo. |
__repr__ |
Method generated by attrs for class MediaVideo. |
__setattr__ |
Method generated by attrs for class MediaVideo. |
Source code in sleap_io/io/video_reading.py
@attrs.define
class MediaVideo(VideoBackend):
"""Video backend for reading videos stored as common media files.
This backend supports reading through FFMPEG (the default), pyav, or OpenCV. Here
are their trade-offs:
- "opencv": Fastest video reader, but only supports a limited number of codecs
and may not be able to read some videos. It requires `opencv-python` to be
installed. It is the fastest because it uses the OpenCV C++ library to read
videos, but is limited by the version of FFMPEG that was linked into it at
build time as well as the OpenCV version used.
- "FFMPEG": Slowest, but most reliable. This is the default backend. It requires
`imageio-ffmpeg` and a `ffmpeg` executable on the system path (which can be
installed via conda). The `imageio` plugin for FFMPEG reads frames into raw
bytes which are communicated to Python through STDOUT on a subprocess pipe,
which can be slow. However, it is the most reliable and feature-complete. If
you install the conda-forge version of ffmpeg, it will be compiled with
support for many codecs, including GPU-accelerated codecs like NVDEC for
H264 and others.
- "pyav": Supports most codecs that FFMPEG does, but not as complete or reliable
of an implementation in `imageio` as FFMPEG for some video types. It is
faster than FFMPEG because it uses the `av` package to read frames directly
into numpy arrays in memory without the need for a subprocess pipe. These
are Python bindings for the C library libav, which is the same library that
FFMPEG uses under the hood.
Attributes:
filename: Path to video file.
grayscale: Whether to force grayscale. If None, autodetect on first frame load.
keep_open: Whether to keep the video reader open between calls to read frames.
If False, will close the reader after each call. If True (the default), it
will keep the reader open and cache it for subsequent calls which may
enhance the performance of reading multiple frames.
plugin: Video plugin to use. One of "opencv", "FFMPEG", or "pyav". If `None`,
will use the first available plugin in the order listed above.
"""
plugin: str = attrs.field()
@plugin.validator
def _validate_plugin(self, attribute, value):
# Normalize the plugin name
normalized = normalize_plugin_name(value)
# Update the actual value to the normalized version
object.__setattr__(self, attribute.name, normalized)
EXTS = ("mp4", "avi", "mov", "mj2", "mkv")
@plugin.default
def _default_plugin(self) -> str:
# Check global default first
if _default_video_plugin is not None:
# Warn if preferred plugin not available
if not _AVAILABLE_VIDEO_BACKENDS.get(_default_video_plugin, False):
import warnings
available = get_available_video_backends()
install_cmd = get_installation_instructions(_default_video_plugin)
warnings.warn(
f"Preferred video plugin '{_default_video_plugin}' is not "
f"available. Available plugins: {available}\n"
f"Install with: {install_cmd}"
)
# Fall through to auto-detection
else:
return _default_video_plugin
# Auto-detect based on what's available
if "cv2" in sys.modules:
return "opencv"
elif "imageio_ffmpeg" in sys.modules:
return "FFMPEG"
elif "av" in sys.modules:
return "pyav"
else:
# Enhanced error message with installation instructions
raise ImportError(
"No video backend plugins are available.\n\n"
"The bundled imageio-ffmpeg should be available by default.\n"
"If you see this error, try reinstalling sleap-io:\n"
" pip install --force-reinstall sleap-io\n\n"
"Alternative backends:\n"
" opencv (fastest): pip install sleap-io[opencv]\n"
" pyav (balanced): pip install sleap-io[pyav]\n\n"
"For more information, see: https://io.sleap.ai"
)
@property
def reader(self) -> object:
"""Return the reader object for the video, caching if necessary."""
if self.keep_open:
if self._open_reader is None:
if self.plugin == "opencv":
self._open_reader = cv2.VideoCapture(self.filename)
elif self.plugin == "pyav" or self.plugin == "FFMPEG":
self._open_reader = iio.imopen(
self.filename, "r", plugin=self.plugin
)
return self._open_reader
else:
if self.plugin == "opencv":
return cv2.VideoCapture(self.filename)
elif self.plugin == "pyav" or self.plugin == "FFMPEG":
return iio.imopen(self.filename, "r", plugin=self.plugin)
@property
def num_frames(self) -> int:
"""Number of frames in the video."""
if self.plugin == "opencv":
return int(self.reader.get(cv2.CAP_PROP_FRAME_COUNT))
else:
props = iio.improps(self.filename, plugin=self.plugin)
n_frames = props.n_images
if np.isinf(n_frames):
legacy_reader = self.reader.legacy_get_reader()
# Note: This might be super slow for some videos, so maybe we should
# defer evaluation of this or give the user control over it.
n_frames = legacy_reader.count_frames()
return n_frames
@property
def fps(self) -> float | None:
"""Frames per second from video container metadata.
Returns:
The FPS from the video container, or None if it cannot be determined.
Notes:
This reads the FPS from the video file metadata using the appropriate
method for the current plugin:
- OpenCV: cv2.CAP_PROP_FPS
- FFMPEG/pyav: imageio metadata
For remote (URL) filenames the FPS is read directly from the pyav
container via ``av.open(url)``. imageio's v2 FFMPEG reader (used for
local files) requires the ``imageio-ffmpeg`` package and an ffmpeg
executable, which are not guaranteed in a pyav-only install, whereas
``av`` is already required for remote loading.
"""
# Return cached/explicit value if set
if self._fps is not None:
return self._fps
# Read from container metadata and cache the result so repeated access
# is O(1). This matters most for the remote (URL) path: ``av.open`` over
# http does not use Range requests, so each uncached read re-streams the
# entire video. Public helpers such as ``Video.frame_to_seconds`` read
# ``fps`` multiple times per call, which would otherwise re-download the
# whole video each time. Container fps is immutable for a given file, and
# the cache slot is cleared when the filename changes (see
# ``Video.replace_filename``), so caching is safe.
try:
if self.plugin == "opencv":
fps = self.reader.get(cv2.CAP_PROP_FPS)
rate = fps if fps > 0 else None
elif _remote._is_url(self.filename):
rate = _fps_from_av_container(self.filename)
else:
# Use imageio v2 API to get metadata (v3 improps doesn't include fps)
import imageio.v2 as iio_v2
reader = iio_v2.get_reader(self.filename, format="FFMPEG")
meta = reader.get_meta_data()
reader.close()
fps = meta.get("fps")
rate = float(fps) if fps is not None else None
except Exception:
return None
self._fps = rate
return rate
@fps.setter
def fps(self, value: float | None) -> None:
"""Set an explicit FPS override.
Args:
value: Frames per second. Must be positive if not None.
Raises:
ValueError: If value is not positive.
Notes:
Setting FPS on MediaVideo overrides the value from container metadata.
This can be useful when the container metadata is incorrect or missing.
"""
if value is not None and value <= 0:
raise ValueError(f"FPS must be positive, got {value}")
self._fps = value
def _read_frame(self, frame_idx: int) -> np.ndarray:
"""Read a single frame from the video.
Args:
frame_idx: Index of frame to read.
Returns:
The frame as a numpy array of shape `(height, width, channels)`.
Notes:
This does not apply grayscale conversion. It is recommended to use the
`get_frame` method of the `VideoBackend` class instead.
"""
if self.plugin == "opencv":
if self.keep_open:
if self._open_reader is None:
self._open_reader = cv2.VideoCapture(self.filename)
reader = self._open_reader
else:
reader = cv2.VideoCapture(self.filename)
if reader.get(cv2.CAP_PROP_POS_FRAMES) != frame_idx:
reader.set(cv2.CAP_PROP_POS_FRAMES, frame_idx)
success, img = reader.read()
if success:
img = img[..., ::-1] # BGR -> RGB
elif self.plugin == "pyav" or self.plugin == "FFMPEG":
if self.keep_open:
img = self.reader.read(index=frame_idx)
else:
with iio.imopen(self.filename, "r", plugin=self.plugin) as reader:
img = reader.read(index=frame_idx)
success = img is not None
if not success:
raise IndexError(f"Failed to read frame index {frame_idx}.")
return img
def _read_frames(self, frame_inds: list) -> np.ndarray:
"""Read a list of frames from the video.
Args:
frame_inds: List of indices of frames to read.
Returns:
The frame as a numpy array of shape `(frames, height, width, channels)`.
Notes:
This does not apply grayscale conversion. It is recommended to use the
`get_frames` method of the `VideoBackend` class instead.
"""
if self.plugin == "opencv":
if self.keep_open:
if self._open_reader is None:
self._open_reader = cv2.VideoCapture(self.filename)
reader = self._open_reader
else:
reader = cv2.VideoCapture(self.filename)
reader.set(cv2.CAP_PROP_POS_FRAMES, frame_inds[0])
imgs = []
for idx in frame_inds:
if reader.get(cv2.CAP_PROP_POS_FRAMES) != idx:
reader.set(cv2.CAP_PROP_POS_FRAMES, idx)
_, img = reader.read()
imgs.append(img)
imgs = np.stack(imgs, axis=0)
imgs = imgs[..., ::-1] # BGR -> RGB
elif self.plugin == "pyav" or self.plugin == "FFMPEG":
if self.keep_open:
if self._open_reader is None:
self._open_reader = iio.imopen(
self.filename, "r", plugin=self.plugin
)
reader = self._open_reader
imgs = np.stack([reader.read(index=idx) for idx in frame_inds], axis=0)
else:
with iio.imopen(self.filename, "r", plugin=self.plugin) as reader:
imgs = np.stack(
[reader.read(index=idx) for idx in frame_inds], axis=0
)
return imgs
EXTS = ('mp4', 'avi', 'mov', 'mj2', 'mkv')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__annotations__ = {'plugin': 'str'}
class-attribute
¶
dict() -> new empty dictionary dict(mapping) -> new dictionary initialized from a mapping object's (key, value) pairs dict(iterable) -> new dictionary initialized as if via: d = {} for k, v in iterable: d[k] = v dict(**kwargs) -> new dictionary initialized with the name=value pairs in the keyword argument list. For example: dict(one=1, two=2)
__attrs_own_setattr__ = True
class-attribute
¶
Returns True when the argument is true, False otherwise. The builtins True and False are the only two instances of the class bool. The class bool is a subclass of the class int, and cannot be subclassed.
__attrs_props__ = ClassProps(is_exception=False, is_slotted=True, has_weakref_slot=True, is_frozen=False, kw_only=<KeywordOnly.NO: 'no'>, collected_fields_by_mro=True, added_init=True, added_repr=True, added_eq=True, added_ordering=False, hashability=<Hashability.UNHASHABLE: 'unhashable'>, added_match_args=True, added_str=False, added_pickling=True, on_setattr_hook=<function pipe.<locals>.wrapped_pipe at 0x7f08471ce840>, field_transformer=None)
class-attribute
¶
Effective class properties as derived from parameters to attr.s() or
define() decorators.
This is the same data structure that attrs uses internally to decide how to construct the final class.
Warning:
This feature is currently **experimental** and is not covered by our
strict backwards-compatibility guarantees.
Attributes:
| Name | Type | Description |
|---|---|---|
is_exception |
bool
|
Whether the class is treated as an exception class. |
is_slotted |
bool
|
Whether the class is |
has_weakref_slot |
bool
|
Whether the class has a slot for weak references. |
is_frozen |
bool
|
Whether the class is frozen. |
kw_only |
KeywordOnly
|
Whether / how the class enforces keyword-only arguments on the
|
collected_fields_by_mro |
bool
|
Whether the class fields were collected by method resolution order.
That is, correctly but unlike |
added_init |
bool
|
Whether the class has an attrs-generated |
added_repr |
bool
|
Whether the class has an attrs-generated |
added_eq |
bool
|
Whether the class has attrs-generated equality methods. |
added_ordering |
bool
|
Whether the class has attrs-generated ordering methods. |
hashability |
Hashability
|
How |
added_match_args |
bool
|
Whether the class supports positional |
added_str |
bool
|
Whether the class has an attrs-generated |
added_pickling |
bool
|
Whether the class has attrs-generated |
on_setattr_hook |
Callable[[Any, Attribute[Any], Any], Any] | None
|
The class's |
field_transformer |
Callable[[Attribute[Any]], Attribute[Any]] | None
|
The class's |
.. versionadded:: 25.4.0
__doc__ = 'Video backend for reading videos stored as common media files.\n\nThis backend supports reading through FFMPEG (the default), pyav, or OpenCV. Here\nare their trade-offs:\n\n - "opencv": Fastest video reader, but only supports a limited number of codecs\n and may not be able to read some videos. It requires `opencv-python` to be\n installed. It is the fastest because it uses the OpenCV C++ library to read\n videos, but is limited by the version of FFMPEG that was linked into it at\n build time as well as the OpenCV version used.\n - "FFMPEG": Slowest, but most reliable. This is the default backend. It requires\n `imageio-ffmpeg` and a `ffmpeg` executable on the system path (which can be\n installed via conda). The `imageio` plugin for FFMPEG reads frames into raw\n bytes which are communicated to Python through STDOUT on a subprocess pipe,\n which can be slow. However, it is the most reliable and feature-complete. If\n you install the conda-forge version of ffmpeg, it will be compiled with\n support for many codecs, including GPU-accelerated codecs like NVDEC for\n H264 and others.\n - "pyav": Supports most codecs that FFMPEG does, but not as complete or reliable\n of an implementation in `imageio` as FFMPEG for some video types. It is\n faster than FFMPEG because it uses the `av` package to read frames directly\n into numpy arrays in memory without the need for a subprocess pipe. These\n are Python bindings for the C library libav, which is the same library that\n FFMPEG uses under the hood.\n\nAttributes:\n filename: Path to video file.\n grayscale: Whether to force grayscale. If None, autodetect on first frame load.\n keep_open: Whether to keep the video reader open between calls to read frames.\n If False, will close the reader after each call. If True (the default), it\n will keep the reader open and cache it for subsequent calls which may\n enhance the performance of reading multiple frames.\n plugin: Video plugin to use. One of "opencv", "FFMPEG", or "pyav". If `None`,\n will use the first available plugin in the order listed above.\n'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__firstlineno__ = 847
class-attribute
¶
int([x]) -> integer int(x, base=10) -> integer
Convert a number or string to an integer, or return 0 if no arguments are given. If x is a number, return x.int(). For floating-point numbers, this truncates towards zero.
If x is not a number or if base is given, then x must be a string, bytes, or bytearray instance representing an integer literal in the given base. The literal can be preceded by '+' or '-' and be surrounded by whitespace. The base defaults to 10. Valid bases are 0 and 2-36. Base 0 means to interpret the base from the string as an integer iteral.
int('0b100', base=0) 4
__match_args__ = ('filename', 'grayscale', 'keep_open', '_cached_shape', '_open_reader', '_fps', 'plugin')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__module__ = 'sleap_io.io.video_reading'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__slots__ = ('plugin',)
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__static_attributes__ = ('_fps', '_open_reader')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
fps
property
¶
Frames per second from video container metadata.
Returns:
| Type | Description |
|---|---|
|
The FPS from the video container, or None if it cannot be determined. |
Notes
This reads the FPS from the video file metadata using the appropriate method for the current plugin: - OpenCV: cv2.CAP_PROP_FPS - FFMPEG/pyav: imageio metadata
For remote (URL) filenames the FPS is read directly from the pyav
container via av.open(url). imageio's v2 FFMPEG reader (used for
local files) requires the imageio-ffmpeg package and an ffmpeg
executable, which are not guaranteed in a pyav-only install, whereas
av is already required for remote loading.
num_frames
property
¶
Number of frames in the video.
reader
property
¶
Return the reader object for the video, caching if necessary.
__eq__(other)
¶
Method generated by attrs for class MediaVideo.
__init__(filename, grayscale=None, keep_open=True, cached_shape=None, open_reader=None, fps=None, plugin=NOTHING)
¶
Method generated by attrs for class MediaVideo.
__repr__()
¶
Method generated by attrs for class MediaVideo.
__setattr__(name, val)
¶
Method generated by attrs for class MediaVideo.
Source code in sleap_io/io/video_reading.py
multiple threads are safe: although all reads share one cached fsspec
file-like (a single byte position), h5py serializes every HDF5 C-library
call under a global recursive lock (`h5py._objects.phil`), so the
seek+read pair a frame read performs is never interleaved across threads.
For true read *parallelism* (rather than just safety), construct
independent `Video`/`HDF5Video` instances per worker; each gets its own
fsspec file and block cache.
"""
TiffVideo
¶
Bases: sleap_io.io.video_reading.VideoBackend
Video backend for reading multi-page TIFF stacks.
This backend supports reading multi-page TIFF files as video sequences. Each page in the TIFF is treated as a frame.
Attributes:
| Name | Type | Description |
|---|---|---|
filename |
Path to the multi-page TIFF file. |
|
grayscale |
Whether to force grayscale. If None, autodetect on first frame load. |
|
keep_open |
Whether to keep the reader open between calls to read frames. |
|
format |
Format of the TIFF file ("multi_page", "THW", "HWT", "THWC", "CHWT"). |
Methods:
| Name | Description |
|---|---|
__attrs_post_init__ |
Initialize format if not provided. |
__eq__ |
Method generated by attrs for class TiffVideo. |
__init__ |
Method generated by attrs for class TiffVideo. |
__repr__ |
Method generated by attrs for class TiffVideo. |
detect_format |
Detect TIFF format and shape for single files. |
is_multipage |
Check if a TIFF file contains multiple pages. |
Source code in sleap_io/io/video_reading.py
@attrs.define
class TiffVideo(VideoBackend):
"""Video backend for reading multi-page TIFF stacks.
This backend supports reading multi-page TIFF files as video sequences.
Each page in the TIFF is treated as a frame.
Attributes:
filename: Path to the multi-page TIFF file.
grayscale: Whether to force grayscale. If None, autodetect on first frame load.
keep_open: Whether to keep the reader open between calls to read frames.
format: Format of the TIFF file ("multi_page", "THW", "HWT", "THWC", "CHWT").
"""
EXTS = ("tif", "tiff")
format: str | None = None
@staticmethod
def is_multipage(filename: str) -> bool:
"""Check if a TIFF file contains multiple pages.
Args:
filename: Path to the TIFF file.
Returns:
True if the TIFF contains multiple pages, False otherwise.
"""
try:
# Try to read the second frame
iio.imread(filename, index=1)
return True
except (IndexError, ValueError):
return False
except Exception:
# For any other error, assume it's not multi-page
return False
@staticmethod
def detect_format(filename: str) -> tuple[str, dict]:
"""Detect TIFF format and shape for single files.
Args:
filename: Path to the TIFF file.
Returns:
Tuple of (format_type, metadata) where:
- format_type: "single_frame", "multi_page", "rank3_video", or "rank4_video"
- metadata: dict with shape info and inferred format
"""
try:
# Read first frame to check shape
img = iio.imread(filename, index=0)
shape = img.shape
# Check if multi-page first
is_multi = TiffVideo.is_multipage(filename)
if is_multi:
return "multi_page", {"shape": shape}
# Single page cases
if img.ndim == 2:
# Rank-2: single channel image
return "single_frame", {"shape": shape}
elif img.ndim == 3:
# Rank-3: could be HWC (single frame) or THW/HWT (video)
return TiffVideo._detect_rank3_format(shape)
elif img.ndim == 4:
# Rank-4: video with channels
return TiffVideo._detect_rank4_format(shape)
else:
return "single_frame", {"shape": shape}
except Exception:
return "single_frame", {"shape": None}
@staticmethod
def _detect_rank3_format(shape: tuple) -> tuple[str, dict]:
"""Detect format for rank-3 TIFF files.
Args:
shape: Shape tuple (dim1, dim2, dim3)
Returns:
Tuple of (format_type, metadata)
"""
dim1, dim2, dim3 = shape
# If last dimension is 1 or 3, likely HWC (single frame)
if dim3 in (1, 3):
return "single_frame", {"shape": shape, "format": "HWC"}
# If first two dims are equal, it's likely HWT format
# (most common case for square frames stored as H x W x T)
if dim1 == dim2:
# Default to HWT format for square frames
return "rank3_video", {
"shape": shape,
"format": "HWT",
"height": dim1,
"width": dim2,
"n_frames": dim3,
}
else:
# For non-square frames, check if it could be THW
# This is less common but possible
if dim2 == dim3:
# Could be THW format
return "rank3_video", {
"shape": shape,
"format": "THW",
"n_frames": dim1,
"height": dim2,
"width": dim3,
}
else:
# Default to HWT format
return "rank3_video", {
"shape": shape,
"format": "HWT",
"height": dim1,
"width": dim2,
"n_frames": dim3,
}
@staticmethod
def _detect_rank4_format(shape: tuple) -> tuple[str, dict]:
"""Detect format for rank-4 TIFF files.
Args:
shape: Shape tuple (dim1, dim2, dim3, dim4)
Returns:
Tuple of (format_type, metadata)
"""
dim1, dim2, dim3, dim4 = shape
# Check if first or last dimension is 1 or 3 (channels)
if dim1 in (1, 3):
# CHWT format
return "rank4_video", {
"shape": shape,
"format": "CHWT",
"channels": dim1,
"height": dim2,
"width": dim3,
"n_frames": dim4,
}
elif dim4 in (1, 3):
# THWC format
return "rank4_video", {
"shape": shape,
"format": "THWC",
"n_frames": dim1,
"height": dim2,
"width": dim3,
"channels": dim4,
}
else:
# Default to THWC
return "rank4_video", {
"shape": shape,
"format": "THWC",
"n_frames": dim1,
"height": dim2,
"width": dim3,
"channels": dim4,
}
def __attrs_post_init__(self):
"""Initialize format if not provided."""
if self.format is None:
# Auto-detect format
format_type, metadata = TiffVideo.detect_format(self.filename)
if format_type == "multi_page":
self.format = "multi_page"
elif format_type in ("rank3_video", "rank4_video"):
self.format = metadata.get("format", "multi_page")
else:
self.format = "multi_page"
@property
def num_frames(self) -> int:
"""Number of frames in the TIFF stack."""
if self.format == "multi_page":
# Count frames by trying to read each one until we get an error
frame_count = 0
while True:
try:
iio.imread(self.filename, index=frame_count)
frame_count += 1
except (IndexError, ValueError):
break
return frame_count
else:
# For rank3/rank4 formats, detect from shape
format_type, metadata = TiffVideo.detect_format(self.filename)
return metadata.get("n_frames", 1)
def _read_frame(self, frame_idx: int) -> np.ndarray:
"""Read a single frame from the TIFF stack.
Args:
frame_idx: Index of frame to read.
Returns:
The frame as a numpy array of shape `(height, width, channels)`.
Notes:
This does not apply grayscale conversion. It is recommended to use the
`get_frame` method of the `VideoBackend` class instead.
"""
if self.format == "multi_page":
img = iio.imread(self.filename, index=frame_idx)
if img.ndim == 2:
img = np.expand_dims(img, axis=-1)
return img
else:
# Read entire array for rank3/rank4 formats
img = iio.imread(self.filename)
if self.format == "THW":
# Extract frame from THW format
frame = img[frame_idx, :, :]
return np.expand_dims(frame, axis=-1)
elif self.format == "HWT":
# Extract frame from HWT format
frame = img[:, :, frame_idx]
return np.expand_dims(frame, axis=-1)
elif self.format == "THWC":
# Extract frame from THWC format
return img[frame_idx, :, :, :]
elif self.format == "CHWT":
# Extract frame from CHWT format
frame = img[:, :, :, frame_idx]
return np.moveaxis(frame, 0, -1) # CHW -> HWC
else:
raise ValueError(f"Unknown format: {self.format}")
def _read_frames(self, frame_inds: list) -> np.ndarray:
"""Read multiple frames from the TIFF stack.
Args:
frame_inds: List of frame indices to read.
Returns:
Frames as a numpy array of shape `(frames, height, width, channels)`.
"""
if self.format == "multi_page":
imgs = []
for idx in frame_inds:
imgs.append(self._read_frame(idx))
return np.stack(imgs, axis=0)
else:
# For rank3/rank4, read all at once and extract
img = iio.imread(self.filename)
if self.format == "THW":
frames = img[frame_inds, :, :]
return np.expand_dims(frames, axis=-1)
elif self.format == "HWT":
frames = img[:, :, frame_inds]
frames = np.moveaxis(frames, -1, 0) # HWT -> THW
return np.expand_dims(frames, axis=-1)
elif self.format == "THWC":
return img[frame_inds, :, :, :]
elif self.format == "CHWT":
frames = img[:, :, :, frame_inds]
frames = np.moveaxis(frames, -1, 0) # CHWT -> TCHW
frames = np.moveaxis(frames, 1, -1) # TCHW -> THWC
return frames
else:
raise ValueError(f"Unknown format: {self.format}")
EXTS = ('tif', 'tiff')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__annotations__ = {'format': 'str | None'}
class-attribute
¶
dict() -> new empty dictionary dict(mapping) -> new dictionary initialized from a mapping object's (key, value) pairs dict(iterable) -> new dictionary initialized as if via: d = {} for k, v in iterable: d[k] = v dict(**kwargs) -> new dictionary initialized with the name=value pairs in the keyword argument list. For example: dict(one=1, two=2)
__attrs_own_setattr__ = False
class-attribute
¶
Returns True when the argument is true, False otherwise. The builtins True and False are the only two instances of the class bool. The class bool is a subclass of the class int, and cannot be subclassed.
__attrs_props__ = ClassProps(is_exception=False, is_slotted=True, has_weakref_slot=True, is_frozen=False, kw_only=<KeywordOnly.NO: 'no'>, collected_fields_by_mro=True, added_init=True, added_repr=True, added_eq=True, added_ordering=False, hashability=<Hashability.UNHASHABLE: 'unhashable'>, added_match_args=True, added_str=False, added_pickling=True, on_setattr_hook=<function pipe.<locals>.wrapped_pipe at 0x7f08471ce840>, field_transformer=None)
class-attribute
¶
Effective class properties as derived from parameters to attr.s() or
define() decorators.
This is the same data structure that attrs uses internally to decide how to construct the final class.
Warning:
This feature is currently **experimental** and is not covered by our
strict backwards-compatibility guarantees.
Attributes:
| Name | Type | Description |
|---|---|---|
is_exception |
bool
|
Whether the class is treated as an exception class. |
is_slotted |
bool
|
Whether the class is |
has_weakref_slot |
bool
|
Whether the class has a slot for weak references. |
is_frozen |
bool
|
Whether the class is frozen. |
kw_only |
KeywordOnly
|
Whether / how the class enforces keyword-only arguments on the
|
collected_fields_by_mro |
bool
|
Whether the class fields were collected by method resolution order.
That is, correctly but unlike |
added_init |
bool
|
Whether the class has an attrs-generated |
added_repr |
bool
|
Whether the class has an attrs-generated |
added_eq |
bool
|
Whether the class has attrs-generated equality methods. |
added_ordering |
bool
|
Whether the class has attrs-generated ordering methods. |
hashability |
Hashability
|
How |
added_match_args |
bool
|
Whether the class supports positional |
added_str |
bool
|
Whether the class has an attrs-generated |
added_pickling |
bool
|
Whether the class has attrs-generated |
on_setattr_hook |
Callable[[Any, Attribute[Any], Any], Any] | None
|
The class's |
field_transformer |
Callable[[Attribute[Any]], Attribute[Any]] | None
|
The class's |
.. versionadded:: 25.4.0
__doc__ = 'Video backend for reading multi-page TIFF stacks.\n\nThis backend supports reading multi-page TIFF files as video sequences.\nEach page in the TIFF is treated as a frame.\n\nAttributes:\n filename: Path to the multi-page TIFF file.\n grayscale: Whether to force grayscale. If None, autodetect on first frame load.\n keep_open: Whether to keep the reader open between calls to read frames.\n format: Format of the TIFF file ("multi_page", "THW", "HWT", "THWC", "CHWT").\n'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__firstlineno__ = 2007
class-attribute
¶
int([x]) -> integer int(x, base=10) -> integer
Convert a number or string to an integer, or return 0 if no arguments are given. If x is a number, return x.int(). For floating-point numbers, this truncates towards zero.
If x is not a number or if base is given, then x must be a string, bytes, or bytearray instance representing an integer literal in the given base. The literal can be preceded by '+' or '-' and be surrounded by whitespace. The base defaults to 10. Valid bases are 0 and 2-36. Base 0 means to interpret the base from the string as an integer iteral.
int('0b100', base=0) 4
__match_args__ = ('filename', 'grayscale', 'keep_open', '_cached_shape', '_open_reader', '_fps', 'format')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__module__ = 'sleap_io.io.video_reading'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__slots__ = ('format',)
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__static_attributes__ = ('format',)
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
num_frames
property
¶
Number of frames in the TIFF stack.
__attrs_post_init__()
¶
Initialize format if not provided.
Source code in sleap_io/io/video_reading.py
def __attrs_post_init__(self):
"""Initialize format if not provided."""
if self.format is None:
# Auto-detect format
format_type, metadata = TiffVideo.detect_format(self.filename)
if format_type == "multi_page":
self.format = "multi_page"
elif format_type in ("rank3_video", "rank4_video"):
self.format = metadata.get("format", "multi_page")
else:
self.format = "multi_page"
__eq__(other)
¶
Method generated by attrs for class TiffVideo.
__init__(filename, grayscale=None, keep_open=True, cached_shape=None, open_reader=None, fps=None, format=None)
¶
__repr__()
¶
Method generated by attrs for class TiffVideo.
detect_format(filename)
staticmethod
¶
Detect TIFF format and shape for single files.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
filename
|
str
|
Path to the TIFF file. |
required |
Returns:
| Type | Description |
|---|---|
tuple[str, dict]
|
Tuple of (format_type, metadata) where: - format_type: "single_frame", "multi_page", "rank3_video", or "rank4_video" - metadata: dict with shape info and inferred format |
Source code in sleap_io/io/video_reading.py
@staticmethod
def detect_format(filename: str) -> tuple[str, dict]:
"""Detect TIFF format and shape for single files.
Args:
filename: Path to the TIFF file.
Returns:
Tuple of (format_type, metadata) where:
- format_type: "single_frame", "multi_page", "rank3_video", or "rank4_video"
- metadata: dict with shape info and inferred format
"""
try:
# Read first frame to check shape
img = iio.imread(filename, index=0)
shape = img.shape
# Check if multi-page first
is_multi = TiffVideo.is_multipage(filename)
if is_multi:
return "multi_page", {"shape": shape}
# Single page cases
if img.ndim == 2:
# Rank-2: single channel image
return "single_frame", {"shape": shape}
elif img.ndim == 3:
# Rank-3: could be HWC (single frame) or THW/HWT (video)
return TiffVideo._detect_rank3_format(shape)
elif img.ndim == 4:
# Rank-4: video with channels
return TiffVideo._detect_rank4_format(shape)
else:
return "single_frame", {"shape": shape}
except Exception:
return "single_frame", {"shape": None}
is_multipage(filename)
staticmethod
¶
Check if a TIFF file contains multiple pages.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
filename
|
str
|
Path to the TIFF file. |
required |
Returns:
| Type | Description |
|---|---|
bool
|
True if the TIFF contains multiple pages, False otherwise. |
Source code in sleap_io/io/video_reading.py
@staticmethod
def is_multipage(filename: str) -> bool:
"""Check if a TIFF file contains multiple pages.
Args:
filename: Path to the TIFF file.
Returns:
True if the TIFF contains multiple pages, False otherwise.
"""
try:
# Try to read the second frame
iio.imread(filename, index=1)
return True
except (IndexError, ValueError):
return False
except Exception:
# For any other error, assume it's not multi-page
return False
VideoBackend
¶
Base class for video backends.
This class is not meant to be used directly. Instead, use the from_filename
constructor to create a backend instance.
Attributes:
| Name | Type | Description |
|---|---|---|
filename |
Path to video file(s). |
|
grayscale |
Whether to force grayscale. If None, autodetect on first frame load. |
|
keep_open |
Whether to keep the video reader open between calls to read frames. If False, will close the reader after each call. If True (the default), it will keep the reader open and cache it for subsequent calls which may enhance the performance of reading multiple frames. |
|
fps |
Frames per second of the video. For MediaVideo, this is read from container metadata. For other backends (ImageVideo, HDF5Video, TiffVideo), this must be set explicitly or will be None. |
Methods:
| Name | Description |
|---|---|
__eq__ |
Method generated by attrs for class VideoBackend. |
__getitem__ |
Return a single frame or a list of frames from the video. |
__getstate__ |
Return state for pickling/deepcopy, dropping the open reader handle. |
__init__ |
Method generated by attrs for class VideoBackend. |
__len__ |
Return number of frames in the video. |
__repr__ |
Method generated by attrs for class VideoBackend. |
__setstate__ |
Restore state from pickling/deepcopy. |
close |
Release the cached open reader handle, if any. |
detect_grayscale |
Detect whether the video is grayscale. |
from_filename |
Create a VideoBackend from a filename. |
get_frame |
Read a single frame from the video. |
get_frames |
Read a list of frames from the video. |
has_frame |
Check if a frame index is contained in the video. |
read_test_frame |
Read a single frame from the video to test for grayscale. |
Source code in sleap_io/io/video_reading.py
@attrs.define
class VideoBackend:
"""Base class for video backends.
This class is not meant to be used directly. Instead, use the `from_filename`
constructor to create a backend instance.
Attributes:
filename: Path to video file(s).
grayscale: Whether to force grayscale. If None, autodetect on first frame load.
keep_open: Whether to keep the video reader open between calls to read frames.
If False, will close the reader after each call. If True (the default), it
will keep the reader open and cache it for subsequent calls which may
enhance the performance of reading multiple frames.
fps: Frames per second of the video. For MediaVideo, this is read from container
metadata. For other backends (ImageVideo, HDF5Video, TiffVideo), this must
be set explicitly or will be None.
"""
filename: str | Path | list[str] | list[Path]
grayscale: bool | None = None
keep_open: bool = True
_cached_shape: tuple[int, int, int, int] | None = None
_open_reader: object | None = None
_fps: float | None = None
@property
def fps(self) -> float | None:
"""Frames per second of the video.
Returns:
The FPS if known, or None if unavailable/unknown.
Notes:
For MediaVideo, this is read from container metadata.
For ImageVideo, HDF5Video, and TiffVideo, this must be set explicitly
or inherited from source_video.
"""
return self._fps
@fps.setter
def fps(self, value: float | None) -> None:
"""Set the FPS.
Args:
value: Frames per second. Must be positive if not None.
Raises:
ValueError: If value is not positive.
"""
if value is not None and value <= 0:
raise ValueError(f"FPS must be positive, got {value}")
self._fps = value
def __getstate__(self) -> dict:
"""Return state for pickling/deepcopy, dropping the open reader handle.
The cached ``_open_reader`` (e.g. an ``h5py.File`` or video container) is
not picklable and is reopened lazily on next access, so it is excluded.
"""
import attr
state = {a.name: getattr(self, a.name) for a in attr.fields(type(self))}
state["_open_reader"] = None
return state
def __setstate__(self, state: dict) -> None:
"""Restore state from pickling/deepcopy.
attrs slotted classes need ``object.__setattr__`` to set slots directly.
Validators are skipped, which is safe since state came from a valid object.
"""
for key, value in state.items():
object.__setattr__(self, key, value)
def close(self) -> None:
"""Release the cached open reader handle, if any.
Closes (``.close()``) or releases (``.release()`` for an OpenCV
``VideoCapture``) the cached ``_open_reader`` and drops the reference so
a long-lived backend does not leak the underlying file/container handle.
The reader is lazily reopened on the next read, so this is safe to call
between reads. A no-op when nothing is cached. Subclasses that hold
additional handles (e.g. :class:`HDF5Video`'s URL file-like) override
this and call ``super().close()``.
"""
reader = self._open_reader
self._open_reader = None
if reader is None:
return
# Every real reader is an h5py.File / imageio reader (``.close()``) or an
# OpenCV VideoCapture (``.release()``); the None case is purely defensive.
closer = getattr(reader, "close", None) or getattr(reader, "release", None)
if closer is None: # pragma: no cover - defensive: reader always closeable
return
try:
closer()
except Exception: # pragma: no cover - defensive: close should not raise
pass
@classmethod
def from_filename(
cls,
filename: str | list[str],
dataset: str | None = None,
grayscale: bool | None = None,
keep_open: bool = True,
url_headers: dict[str, str] | None = None,
url_stream_mode: str = "blockcache",
**kwargs,
) -> "VideoBackend":
"""Create a VideoBackend from a filename.
Args:
filename: Path to video file(s).
dataset: Name of dataset in HDF5 file.
grayscale: Whether to force grayscale. If None, autodetect on first frame
load.
keep_open: Whether to keep the video reader open between calls to read
frames. If False, will close the reader after each call. If True (the
default), it will keep the reader open and cache it for subsequent calls
which may enhance the performance of reading multiple frames.
url_headers: HTTP headers forwarded to the remote backend when
``filename`` is a URL (HDF5Video only). Set at construction so the
metadata probe is authenticated; ignored for local files and other
backends.
url_stream_mode: Remote streaming strategy for a URL-backed HDF5Video
(one of ``"blockcache"``/``"cache"``/``"filecache"``/``"download"``).
Ignored for local files and other backends.
**kwargs: Additional backend-specific arguments. These are filtered to only
include parameters that are valid for the specific backend being
created:
- For ImageVideo: plugin (str): Image plugin to use. One of "opencv"
or "imageio". Also accepts aliases (case-insensitive).
If None, uses global default if set, otherwise auto-detects.
- For MediaVideo: plugin (str): Video plugin to use. One of "opencv",
"FFMPEG", or "pyav". Also accepts aliases (case-insensitive).
If None, uses global default if set, otherwise auto-detects.
- For HDF5Video: input_format (str), frame_map (dict),
source_filename (str),
source_inds (np.ndarray), image_format (str). See HDF5Video for
details.
Returns:
VideoBackend subclass instance.
"""
if isinstance(filename, Path):
filename = filename.as_posix()
is_url = type(filename) is str and _remote._is_url(filename)
if is_url:
from sleap_io.io._gdrive import _is_gdrive_url
if _is_gdrive_url(filename):
# Drive download URLs carry no extension and Drive rejects the
# range/HEAD requests video decoding relies on, so streaming a
# Drive video is not supported. Drive *labels* (.slp) loading is
# supported via load_slp/load_file.
raise NotImplementedError(
"Loading videos directly from Google Drive URLs is not "
"supported (Drive download links carry no file extension and "
"reject the range requests video decoding needs). Download "
"the video file first, or load Drive .slp label files with "
f"load_slp/load_file. (URL: {_remote._redact_url(filename)})"
)
# Skip local-filesystem dir detection for URLs (``Path.is_dir`` on a URL
# is meaningless and would just return False, but avoid the syscall).
if type(filename) is str and not is_url and Path(filename).is_dir():
filename = ImageVideo.find_images(filename)
# Match extensions against the URL *path* (query/fragment stripped) for
# URLs, and the lowercased filename otherwise.
ext_token = _extension_token(filename) if type(filename) is str else ""
if type(filename) is list:
filename = [Path(f).as_posix() for f in filename]
return ImageVideo(
filename, grayscale=grayscale, **_get_valid_kwargs(ImageVideo, kwargs)
)
elif ext_token.endswith(("tif", "tiff")):
# Detect TIFF format
format_type, metadata = TiffVideo.detect_format(filename)
if format_type in ("multi_page", "rank3_video", "rank4_video"):
# Use TiffVideo for multi-page or multi-dimensional TIFFs
tiff_kwargs = _get_valid_kwargs(TiffVideo, kwargs)
# Add format if detected
if format_type in ("rank3_video", "rank4_video"):
tiff_kwargs["format"] = metadata.get("format")
return TiffVideo(
filename,
grayscale=grayscale,
keep_open=keep_open,
**tiff_kwargs,
)
else:
# Single-page TIFF, treat as regular image
return ImageVideo(
[filename],
grayscale=grayscale,
**_get_valid_kwargs(ImageVideo, kwargs),
)
elif ext_token.endswith(tuple(ext.lower() for ext in ImageVideo.EXTS)):
return ImageVideo(
[filename], grayscale=grayscale, **_get_valid_kwargs(ImageVideo, kwargs)
)
elif ext_token.endswith(".seq"):
from sleap_io.io.seq import SeqVideo
return SeqVideo(
filename,
grayscale=grayscale,
keep_open=keep_open,
**_get_valid_kwargs(SeqVideo, kwargs),
)
elif ext_token.endswith(tuple(ext.lower() for ext in MediaVideo.EXTS)):
media_kwargs = _get_valid_kwargs(MediaVideo, kwargs)
if is_url:
# Remote media videos are read via pyav (imageio's pyav plugin
# forwards http(s) URIs to ``av.open`` natively). Enforce the
# documented contract that only http/https URLs are supported:
# cloud schemes (s3/gs/gcs/az/abfs) are recognized as remote by
# ``_is_url`` but are not safe to hand to ``av.open``, so reject
# them cleanly here rather than letting the raw URL reach the
# decoder.
scheme = urllib.parse.urlparse(filename).scheme.lower()
if scheme not in ("http", "https"):
raise NotImplementedError(
"Remote video loading only supports http/https URLs; "
f"got scheme '{scheme}' for "
f"{_remote._redact_url(filename)}. Download the file "
"locally first."
)
# Remote media video is decoded by handing the raw URL to
# ``av.open`` (via imageio's pyav plugin), which has no hook for
# forwarding HTTP request headers or selecting an fsspec stream
# mode. Auth/streaming kwargs that work for remote .slp/.pkg.slp
# (HDF5Video) therefore cannot be honored here. Rather than
# silently drop them and return an unauthenticated backend,
# reject them with an actionable error. ``url_headers`` /
# ``url_stream_mode`` are the explicit ``from_filename``
# parameters; ``headers`` / ``stream_mode`` arrive via
# ``**kwargs`` (e.g. from ``load_video(url, headers=...)``).
if (
url_headers is not None
or url_stream_mode != "blockcache"
or kwargs.get("headers") is not None
or kwargs.get("stream_mode") not in (None, "auto")
):
raise ValueError(
"Remote media video cannot be authenticated with "
"'headers'/'url_headers' or configured with a stream "
"mode: it is decoded by handing the URL directly to "
"FFmpeg (via pyav), which does not support custom HTTP "
"headers or fsspec streaming. Use a pre-signed URL that "
"embeds credentials in the query string, or download "
"the file locally first. (These options do work for "
"remote .slp/.pkg.slp labels.) (URL: "
f"{_remote._redact_url(filename)})"
)
# Default to pyav when the caller did not request a specific
# plugin, and require the ``av`` package up front for a clear
# error.
if media_kwargs.get("plugin") is None:
media_kwargs["plugin"] = "pyav"
if (
normalize_plugin_name(media_kwargs["plugin"]) == "pyav"
and not _is_pyav_available()
):
# Defensive: ``av`` is required for remote loading and is
# always present in the test/CI environment, so this guard
# only fires for an install lacking the ``[pyav]`` extra.
raise ImportError( # pragma: no cover
"Loading videos from URLs requires the 'av' package "
"(pyav). Install with: pip install 'sleap-io[pyav]'. "
f"(URL: {_remote._redact_url(filename)})"
)
return MediaVideo(
filename,
grayscale=grayscale,
keep_open=keep_open,
**media_kwargs,
)
elif ext_token.endswith(tuple(ext.lower() for ext in HDF5Video.EXTS)):
# Pass ``url_headers`` / ``url_stream_mode`` explicitly (not via
# ``_get_valid_kwargs``, which keys on the underscored field *name*
# and would drop the alias) so the construction-time probe in
# ``HDF5Video.__attrs_post_init__`` is authenticated for remote URLs.
return HDF5Video(
filename,
dataset=dataset,
grayscale=grayscale,
keep_open=keep_open,
url_headers=url_headers,
url_stream_mode=url_stream_mode,
**_get_valid_kwargs(HDF5Video, kwargs),
)
else:
raise ValueError(f"Unknown video file type: {filename}")
def _read_frame(self, frame_idx: int) -> np.ndarray:
"""Read a single frame from the video. Must be implemented in subclasses."""
raise NotImplementedError
def _read_frames(self, frame_inds: list) -> np.ndarray:
"""Read a list of frames from the video."""
return np.stack([self.get_frame(i) for i in frame_inds], axis=0)
def read_test_frame(self) -> np.ndarray:
"""Read a single frame from the video to test for grayscale.
Note:
This reads the frame at index 0. This may not be appropriate if the first
frame is not available in a given backend.
"""
return self._read_frame(0)
def detect_grayscale(self, test_img: np.ndarray | None = None) -> bool:
"""Detect whether the video is grayscale.
This works by reading in a test frame and comparing the first and last channel
for equality. It may fail in cases where, due to compression, the first and
last channels are not exactly the same.
Args:
test_img: Optional test image to use. If not provided, a test image will be
loaded via the `read_test_frame` method.
Returns:
Whether the video is grayscale. This value is also cached in the `grayscale`
attribute of the class.
"""
if test_img is None:
test_img = self.read_test_frame()
is_grayscale = np.array_equal(test_img[..., 0], test_img[..., -1])
self.grayscale = is_grayscale
return is_grayscale
@property
def num_frames(self) -> int:
"""Number of frames in the video. Must be implemented in subclasses."""
raise NotImplementedError
@property
def img_shape(self) -> tuple[int, int, int]:
"""Shape of a single frame in the video."""
height, width, channels = self.read_test_frame().shape
if self.grayscale is None:
self.detect_grayscale()
if self.grayscale is False:
channels = 3
elif self.grayscale is True:
channels = 1
return int(height), int(width), int(channels)
@property
def shape(self) -> tuple[int, int, int, int]:
"""Shape of the video as a tuple of `(frames, height, width, channels)`.
On first call, this will defer to `num_frames` and `img_shape` to determine the
full shape. This call may be expensive for some subclasses, so the result is
cached and returned on subsequent calls.
"""
if self._cached_shape is not None:
return self._cached_shape
else:
shape = (self.num_frames,) + self.img_shape
self._cached_shape = shape
return shape
@property
def frames(self) -> int:
"""Number of frames in the video."""
return self.shape[0]
def __len__(self) -> int:
"""Return number of frames in the video."""
return self.shape[0]
def has_frame(self, frame_idx: int) -> bool:
"""Check if a frame index is contained in the video.
Args:
frame_idx: Index of frame to check.
Returns:
`True` if the index is contained in the video, otherwise `False`.
"""
return frame_idx < len(self)
def get_frame(self, frame_idx: int) -> np.ndarray:
"""Read a single frame from the video.
Args:
frame_idx: Index of frame to read.
Returns:
Frame as a numpy array of shape `(height, width, channels)` where the
`channels` dimension is 1 for grayscale videos and 3 for color videos.
Notes:
If the `grayscale` attribute is set to `True`, the `channels` dimension will
be reduced to 1 if an RGB frame is loaded from the backend.
If the `grayscale` attribute is set to `None`, the `grayscale` attribute
will be automatically set based on the first frame read.
See also: `get_frames`
"""
if not self.has_frame(frame_idx):
raise IndexError(f"Frame index {frame_idx} out of range.")
img = self._read_frame(frame_idx)
if self.grayscale is None:
self.detect_grayscale(img)
if self.grayscale:
img = img[..., [0]]
return img
def get_frames(self, frame_inds: list[int]) -> np.ndarray:
"""Read a list of frames from the video.
Depending on the backend implementation, this may be faster than reading frames
individually using `get_frame`.
Args:
frame_inds: List of frame indices to read.
Returns:
Frames as a numpy array of shape `(frames, height, width, channels)` where
`channels` dimension is 1 for grayscale videos and 3 for color videos.
Notes:
If the `grayscale` attribute is set to `True`, the `channels` dimension will
be reduced to 1 if an RGB frame is loaded from the backend.
If the `grayscale` attribute is set to `None`, the `grayscale` attribute
will be automatically set based on the first frame read.
See also: `get_frame`
"""
imgs = self._read_frames(frame_inds)
if self.grayscale is None:
self.detect_grayscale(imgs[0])
if self.grayscale:
imgs = imgs[..., [0]]
return imgs
def __getitem__(self, ind: int | list[int] | slice) -> np.ndarray:
"""Return a single frame or a list of frames from the video.
Args:
ind: Index or list of indices of frames to read.
Returns:
Frame or frames as a numpy array of shape `(height, width, channels)` if a
scalar index is provided, or `(frames, height, width, channels)` if a list
of indices is provided.
See also: get_frame, get_frames
"""
if np.isscalar(ind):
return self.get_frame(ind)
else:
if type(ind) is slice:
start = (ind.start or 0) % len(self)
stop = ind.stop or len(self)
if stop < 0:
stop = len(self) + stop
step = ind.step or 1
ind = range(start, stop, step)
return self.get_frames(ind)
__annotations__ = {'filename': 'str | Path | list[str] | list[Path]', 'grayscale': 'bool | None', 'keep_open': 'bool', '_cached_shape': 'tuple[int, int, int, int] | None', '_open_reader': 'object | None', '_fps': 'float | None'}
class-attribute
¶
dict() -> new empty dictionary dict(mapping) -> new dictionary initialized from a mapping object's (key, value) pairs dict(iterable) -> new dictionary initialized as if via: d = {} for k, v in iterable: d[k] = v dict(**kwargs) -> new dictionary initialized with the name=value pairs in the keyword argument list. For example: dict(one=1, two=2)
__attrs_own_setattr__ = False
class-attribute
¶
Returns True when the argument is true, False otherwise. The builtins True and False are the only two instances of the class bool. The class bool is a subclass of the class int, and cannot be subclassed.
__attrs_props__ = ClassProps(is_exception=False, is_slotted=True, has_weakref_slot=True, is_frozen=False, kw_only=<KeywordOnly.NO: 'no'>, collected_fields_by_mro=True, added_init=True, added_repr=True, added_eq=True, added_ordering=False, hashability=<Hashability.UNHASHABLE: 'unhashable'>, added_match_args=True, added_str=False, added_pickling=False, on_setattr_hook=<function pipe.<locals>.wrapped_pipe at 0x7f08471ce840>, field_transformer=None)
class-attribute
¶
Effective class properties as derived from parameters to attr.s() or
define() decorators.
This is the same data structure that attrs uses internally to decide how to construct the final class.
Warning:
This feature is currently **experimental** and is not covered by our
strict backwards-compatibility guarantees.
Attributes:
| Name | Type | Description |
|---|---|---|
is_exception |
bool
|
Whether the class is treated as an exception class. |
is_slotted |
bool
|
Whether the class is |
has_weakref_slot |
bool
|
Whether the class has a slot for weak references. |
is_frozen |
bool
|
Whether the class is frozen. |
kw_only |
KeywordOnly
|
Whether / how the class enforces keyword-only arguments on the
|
collected_fields_by_mro |
bool
|
Whether the class fields were collected by method resolution order.
That is, correctly but unlike |
added_init |
bool
|
Whether the class has an attrs-generated |
added_repr |
bool
|
Whether the class has an attrs-generated |
added_eq |
bool
|
Whether the class has attrs-generated equality methods. |
added_ordering |
bool
|
Whether the class has attrs-generated ordering methods. |
hashability |
Hashability
|
How |
added_match_args |
bool
|
Whether the class supports positional |
added_str |
bool
|
Whether the class has an attrs-generated |
added_pickling |
bool
|
Whether the class has attrs-generated |
on_setattr_hook |
Callable[[Any, Attribute[Any], Any], Any] | None
|
The class's |
field_transformer |
Callable[[Attribute[Any]], Attribute[Any]] | None
|
The class's |
.. versionadded:: 25.4.0
__doc__ = 'Base class for video backends.\n\nThis class is not meant to be used directly. Instead, use the `from_filename`\nconstructor to create a backend instance.\n\nAttributes:\n filename: Path to video file(s).\n grayscale: Whether to force grayscale. If None, autodetect on first frame load.\n keep_open: Whether to keep the video reader open between calls to read frames.\n If False, will close the reader after each call. If True (the default), it\n will keep the reader open and cache it for subsequent calls which may\n enhance the performance of reading multiple frames.\n fps: Frames per second of the video. For MediaVideo, this is read from container\n metadata. For other backends (ImageVideo, HDF5Video, TiffVideo), this must\n be set explicitly or will be None.\n'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__firstlineno__ = 365
class-attribute
¶
int([x]) -> integer int(x, base=10) -> integer
Convert a number or string to an integer, or return 0 if no arguments are given. If x is a number, return x.int(). For floating-point numbers, this truncates towards zero.
If x is not a number or if base is given, then x must be a string, bytes, or bytearray instance representing an integer literal in the given base. The literal can be preceded by '+' or '-' and be surrounded by whitespace. The base defaults to 10. Valid bases are 0 and 2-36. Base 0 means to interpret the base from the string as an integer iteral.
int('0b100', base=0) 4
__match_args__ = ('filename', 'grayscale', 'keep_open', '_cached_shape', '_open_reader', '_fps')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__module__ = 'sleap_io.io.video_reading'
class-attribute
¶
str(object='') -> str str(bytes_or_buffer[, encoding[, errors]]) -> str
Create a new string object from the given object. If encoding or errors is specified, then the object must expose a data buffer that will be decoded using the given encoding and error handler. Otherwise, returns the result of object.str() (if defined) or repr(object). encoding defaults to 'utf-8'. errors defaults to 'strict'.
__slots__ = ('filename', 'grayscale', 'keep_open', '_cached_shape', '_open_reader', '_fps', '__weakref__')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__static_attributes__ = ('_cached_shape', '_fps', '_open_reader', 'grayscale')
class-attribute
¶
Built-in immutable sequence.
If no argument is given, the constructor returns an empty tuple. If iterable is specified the tuple is initialized from iterable's items.
If the argument is a tuple, the return value is the same object.
__weakref__
property
¶
list of weak references to the object
fps
property
¶
Frames per second of the video.
Returns:
| Type | Description |
|---|---|
|
The FPS if known, or None if unavailable/unknown. |
Notes
For MediaVideo, this is read from container metadata. For ImageVideo, HDF5Video, and TiffVideo, this must be set explicitly or inherited from source_video.
frames
property
¶
Number of frames in the video.
img_shape
property
¶
Shape of a single frame in the video.
num_frames
property
¶
Number of frames in the video. Must be implemented in subclasses.
shape
property
¶
Shape of the video as a tuple of (frames, height, width, channels).
On first call, this will defer to num_frames and img_shape to determine the
full shape. This call may be expensive for some subclasses, so the result is
cached and returned on subsequent calls.
__eq__(other)
¶
Method generated by attrs for class VideoBackend.
__getitem__(ind)
¶
Return a single frame or a list of frames from the video.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
ind
|
int | list[int] | slice
|
Index or list of indices of frames to read. |
required |
Returns:
| Type | Description |
|---|---|
ndarray
|
Frame or frames as a numpy array of shape |
See also: get_frame, get_frames
Source code in sleap_io/io/video_reading.py
def __getitem__(self, ind: int | list[int] | slice) -> np.ndarray:
"""Return a single frame or a list of frames from the video.
Args:
ind: Index or list of indices of frames to read.
Returns:
Frame or frames as a numpy array of shape `(height, width, channels)` if a
scalar index is provided, or `(frames, height, width, channels)` if a list
of indices is provided.
See also: get_frame, get_frames
"""
if np.isscalar(ind):
return self.get_frame(ind)
else:
if type(ind) is slice:
start = (ind.start or 0) % len(self)
stop = ind.stop or len(self)
if stop < 0:
stop = len(self) + stop
step = ind.step or 1
ind = range(start, stop, step)
return self.get_frames(ind)
__getstate__()
¶
Return state for pickling/deepcopy, dropping the open reader handle.
The cached _open_reader (e.g. an h5py.File or video container) is
not picklable and is reopened lazily on next access, so it is excluded.
Source code in sleap_io/io/video_reading.py
def __getstate__(self) -> dict:
"""Return state for pickling/deepcopy, dropping the open reader handle.
The cached ``_open_reader`` (e.g. an ``h5py.File`` or video container) is
not picklable and is reopened lazily on next access, so it is excluded.
"""
import attr
state = {a.name: getattr(self, a.name) for a in attr.fields(type(self))}
state["_open_reader"] = None
return state
__init__(filename, grayscale=None, keep_open=True, cached_shape=None, open_reader=None, fps=None)
¶
__len__()
¶
__repr__()
¶
Method generated by attrs for class VideoBackend.
__setstate__(state)
¶
Restore state from pickling/deepcopy.
attrs slotted classes need object.__setattr__ to set slots directly.
Validators are skipped, which is safe since state came from a valid object.
Source code in sleap_io/io/video_reading.py
def __setstate__(self, state: dict) -> None:
"""Restore state from pickling/deepcopy.
attrs slotted classes need ``object.__setattr__`` to set slots directly.
Validators are skipped, which is safe since state came from a valid object.
"""
for key, value in state.items():
object.__setattr__(self, key, value)
close()
¶
Release the cached open reader handle, if any.
Closes (.close()) or releases (.release() for an OpenCV
VideoCapture) the cached _open_reader and drops the reference so
a long-lived backend does not leak the underlying file/container handle.
The reader is lazily reopened on the next read, so this is safe to call
between reads. A no-op when nothing is cached. Subclasses that hold
additional handles (e.g. :class:HDF5Video's URL file-like) override
this and call super().close().
Source code in sleap_io/io/video_reading.py
def close(self) -> None:
"""Release the cached open reader handle, if any.
Closes (``.close()``) or releases (``.release()`` for an OpenCV
``VideoCapture``) the cached ``_open_reader`` and drops the reference so
a long-lived backend does not leak the underlying file/container handle.
The reader is lazily reopened on the next read, so this is safe to call
between reads. A no-op when nothing is cached. Subclasses that hold
additional handles (e.g. :class:`HDF5Video`'s URL file-like) override
this and call ``super().close()``.
"""
reader = self._open_reader
self._open_reader = None
if reader is None:
return
# Every real reader is an h5py.File / imageio reader (``.close()``) or an
# OpenCV VideoCapture (``.release()``); the None case is purely defensive.
closer = getattr(reader, "close", None) or getattr(reader, "release", None)
if closer is None: # pragma: no cover - defensive: reader always closeable
return
try:
closer()
except Exception: # pragma: no cover - defensive: close should not raise
pass
detect_grayscale(test_img=None)
¶
Detect whether the video is grayscale.
This works by reading in a test frame and comparing the first and last channel for equality. It may fail in cases where, due to compression, the first and last channels are not exactly the same.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
test_img
|
ndarray | None
|
Optional test image to use. If not provided, a test image will be
loaded via the |
None
|
Returns:
| Type | Description |
|---|---|
bool
|
Whether the video is grayscale. This value is also cached in the |
Source code in sleap_io/io/video_reading.py
def detect_grayscale(self, test_img: np.ndarray | None = None) -> bool:
"""Detect whether the video is grayscale.
This works by reading in a test frame and comparing the first and last channel
for equality. It may fail in cases where, due to compression, the first and
last channels are not exactly the same.
Args:
test_img: Optional test image to use. If not provided, a test image will be
loaded via the `read_test_frame` method.
Returns:
Whether the video is grayscale. This value is also cached in the `grayscale`
attribute of the class.
"""
if test_img is None:
test_img = self.read_test_frame()
is_grayscale = np.array_equal(test_img[..., 0], test_img[..., -1])
self.grayscale = is_grayscale
return is_grayscale
from_filename(filename, dataset=None, grayscale=None, keep_open=True, url_headers=None, url_stream_mode='blockcache', **kwargs)
classmethod
¶
Create a VideoBackend from a filename.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
filename
|
str | list[str]
|
Path to video file(s). |
required |
dataset
|
str | None
|
Name of dataset in HDF5 file. |
None
|
grayscale
|
bool | None
|
Whether to force grayscale. If None, autodetect on first frame load. |
None
|
keep_open
|
bool
|
Whether to keep the video reader open between calls to read frames. If False, will close the reader after each call. If True (the default), it will keep the reader open and cache it for subsequent calls which may enhance the performance of reading multiple frames. |
True
|
url_headers
|
dict[str, str] | None
|
HTTP headers forwarded to the remote backend when
|
None
|
url_stream_mode
|
str
|
Remote streaming strategy for a URL-backed HDF5Video
(one of |
'blockcache'
|
**kwargs
|
Additional backend-specific arguments. These are filtered to only include parameters that are valid for the specific backend being created: - For ImageVideo: plugin (str): Image plugin to use. One of "opencv" or "imageio". Also accepts aliases (case-insensitive). If None, uses global default if set, otherwise auto-detects. - For MediaVideo: plugin (str): Video plugin to use. One of "opencv", "FFMPEG", or "pyav". Also accepts aliases (case-insensitive). If None, uses global default if set, otherwise auto-detects. - For HDF5Video: input_format (str), frame_map (dict), source_filename (str), source_inds (np.ndarray), image_format (str). See HDF5Video for details. |
required |
Returns:
| Type | Description |
|---|---|
VideoBackend
|
VideoBackend subclass instance. |
Source code in sleap_io/io/video_reading.py
@classmethod
def from_filename(
cls,
filename: str | list[str],
dataset: str | None = None,
grayscale: bool | None = None,
keep_open: bool = True,
url_headers: dict[str, str] | None = None,
url_stream_mode: str = "blockcache",
**kwargs,
) -> "VideoBackend":
"""Create a VideoBackend from a filename.
Args:
filename: Path to video file(s).
dataset: Name of dataset in HDF5 file.
grayscale: Whether to force grayscale. If None, autodetect on first frame
load.
keep_open: Whether to keep the video reader open between calls to read
frames. If False, will close the reader after each call. If True (the
default), it will keep the reader open and cache it for subsequent calls
which may enhance the performance of reading multiple frames.
url_headers: HTTP headers forwarded to the remote backend when
``filename`` is a URL (HDF5Video only). Set at construction so the
metadata probe is authenticated; ignored for local files and other
backends.
url_stream_mode: Remote streaming strategy for a URL-backed HDF5Video
(one of ``"blockcache"``/``"cache"``/``"filecache"``/``"download"``).
Ignored for local files and other backends.
**kwargs: Additional backend-specific arguments. These are filtered to only
include parameters that are valid for the specific backend being
created:
- For ImageVideo: plugin (str): Image plugin to use. One of "opencv"
or "imageio". Also accepts aliases (case-insensitive).
If None, uses global default if set, otherwise auto-detects.
- For MediaVideo: plugin (str): Video plugin to use. One of "opencv",
"FFMPEG", or "pyav". Also accepts aliases (case-insensitive).
If None, uses global default if set, otherwise auto-detects.
- For HDF5Video: input_format (str), frame_map (dict),
source_filename (str),
source_inds (np.ndarray), image_format (str). See HDF5Video for
details.
Returns:
VideoBackend subclass instance.
"""
if isinstance(filename, Path):
filename = filename.as_posix()
is_url = type(filename) is str and _remote._is_url(filename)
if is_url:
from sleap_io.io._gdrive import _is_gdrive_url
if _is_gdrive_url(filename):
# Drive download URLs carry no extension and Drive rejects the
# range/HEAD requests video decoding relies on, so streaming a
# Drive video is not supported. Drive *labels* (.slp) loading is
# supported via load_slp/load_file.
raise NotImplementedError(
"Loading videos directly from Google Drive URLs is not "
"supported (Drive download links carry no file extension and "
"reject the range requests video decoding needs). Download "
"the video file first, or load Drive .slp label files with "
f"load_slp/load_file. (URL: {_remote._redact_url(filename)})"
)
# Skip local-filesystem dir detection for URLs (``Path.is_dir`` on a URL
# is meaningless and would just return False, but avoid the syscall).
if type(filename) is str and not is_url and Path(filename).is_dir():
filename = ImageVideo.find_images(filename)
# Match extensions against the URL *path* (query/fragment stripped) for
# URLs, and the lowercased filename otherwise.
ext_token = _extension_token(filename) if type(filename) is str else ""
if type(filename) is list:
filename = [Path(f).as_posix() for f in filename]
return ImageVideo(
filename, grayscale=grayscale, **_get_valid_kwargs(ImageVideo, kwargs)
)
elif ext_token.endswith(("tif", "tiff")):
# Detect TIFF format
format_type, metadata = TiffVideo.detect_format(filename)
if format_type in ("multi_page", "rank3_video", "rank4_video"):
# Use TiffVideo for multi-page or multi-dimensional TIFFs
tiff_kwargs = _get_valid_kwargs(TiffVideo, kwargs)
# Add format if detected
if format_type in ("rank3_video", "rank4_video"):
tiff_kwargs["format"] = metadata.get("format")
return TiffVideo(
filename,
grayscale=grayscale,
keep_open=keep_open,
**tiff_kwargs,
)
else:
# Single-page TIFF, treat as regular image
return ImageVideo(
[filename],
grayscale=grayscale,
**_get_valid_kwargs(ImageVideo, kwargs),
)
elif ext_token.endswith(tuple(ext.lower() for ext in ImageVideo.EXTS)):
return ImageVideo(
[filename], grayscale=grayscale, **_get_valid_kwargs(ImageVideo, kwargs)
)
elif ext_token.endswith(".seq"):
from sleap_io.io.seq import SeqVideo
return SeqVideo(
filename,
grayscale=grayscale,
keep_open=keep_open,
**_get_valid_kwargs(SeqVideo, kwargs),
)
elif ext_token.endswith(tuple(ext.lower() for ext in MediaVideo.EXTS)):
media_kwargs = _get_valid_kwargs(MediaVideo, kwargs)
if is_url:
# Remote media videos are read via pyav (imageio's pyav plugin
# forwards http(s) URIs to ``av.open`` natively). Enforce the
# documented contract that only http/https URLs are supported:
# cloud schemes (s3/gs/gcs/az/abfs) are recognized as remote by
# ``_is_url`` but are not safe to hand to ``av.open``, so reject
# them cleanly here rather than letting the raw URL reach the
# decoder.
scheme = urllib.parse.urlparse(filename).scheme.lower()
if scheme not in ("http", "https"):
raise NotImplementedError(
"Remote video loading only supports http/https URLs; "
f"got scheme '{scheme}' for "
f"{_remote._redact_url(filename)}. Download the file "
"locally first."
)
# Remote media video is decoded by handing the raw URL to
# ``av.open`` (via imageio's pyav plugin), which has no hook for
# forwarding HTTP request headers or selecting an fsspec stream
# mode. Auth/streaming kwargs that work for remote .slp/.pkg.slp
# (HDF5Video) therefore cannot be honored here. Rather than
# silently drop them and return an unauthenticated backend,
# reject them with an actionable error. ``url_headers`` /
# ``url_stream_mode`` are the explicit ``from_filename``
# parameters; ``headers`` / ``stream_mode`` arrive via
# ``**kwargs`` (e.g. from ``load_video(url, headers=...)``).
if (
url_headers is not None
or url_stream_mode != "blockcache"
or kwargs.get("headers") is not None
or kwargs.get("stream_mode") not in (None, "auto")
):
raise ValueError(
"Remote media video cannot be authenticated with "
"'headers'/'url_headers' or configured with a stream "
"mode: it is decoded by handing the URL directly to "
"FFmpeg (via pyav), which does not support custom HTTP "
"headers or fsspec streaming. Use a pre-signed URL that "
"embeds credentials in the query string, or download "
"the file locally first. (These options do work for "
"remote .slp/.pkg.slp labels.) (URL: "
f"{_remote._redact_url(filename)})"
)
# Default to pyav when the caller did not request a specific
# plugin, and require the ``av`` package up front for a clear
# error.
if media_kwargs.get("plugin") is None:
media_kwargs["plugin"] = "pyav"
if (
normalize_plugin_name(media_kwargs["plugin"]) == "pyav"
and not _is_pyav_available()
):
# Defensive: ``av`` is required for remote loading and is
# always present in the test/CI environment, so this guard
# only fires for an install lacking the ``[pyav]`` extra.
raise ImportError( # pragma: no cover
"Loading videos from URLs requires the 'av' package "
"(pyav). Install with: pip install 'sleap-io[pyav]'. "
f"(URL: {_remote._redact_url(filename)})"
)
return MediaVideo(
filename,
grayscale=grayscale,
keep_open=keep_open,
**media_kwargs,
)
elif ext_token.endswith(tuple(ext.lower() for ext in HDF5Video.EXTS)):
# Pass ``url_headers`` / ``url_stream_mode`` explicitly (not via
# ``_get_valid_kwargs``, which keys on the underscored field *name*
# and would drop the alias) so the construction-time probe in
# ``HDF5Video.__attrs_post_init__`` is authenticated for remote URLs.
return HDF5Video(
filename,
dataset=dataset,
grayscale=grayscale,
keep_open=keep_open,
url_headers=url_headers,
url_stream_mode=url_stream_mode,
**_get_valid_kwargs(HDF5Video, kwargs),
)
else:
raise ValueError(f"Unknown video file type: {filename}")
get_frame(frame_idx)
¶
Read a single frame from the video.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame_idx
|
int
|
Index of frame to read. |
required |
Returns:
| Type | Description |
|---|---|
ndarray
|
Frame as a numpy array of shape |
Notes
If the grayscale attribute is set to True, the channels dimension will
be reduced to 1 if an RGB frame is loaded from the backend.
If the grayscale attribute is set to None, the grayscale attribute
will be automatically set based on the first frame read.
See also: get_frames
Source code in sleap_io/io/video_reading.py
def get_frame(self, frame_idx: int) -> np.ndarray:
"""Read a single frame from the video.
Args:
frame_idx: Index of frame to read.
Returns:
Frame as a numpy array of shape `(height, width, channels)` where the
`channels` dimension is 1 for grayscale videos and 3 for color videos.
Notes:
If the `grayscale` attribute is set to `True`, the `channels` dimension will
be reduced to 1 if an RGB frame is loaded from the backend.
If the `grayscale` attribute is set to `None`, the `grayscale` attribute
will be automatically set based on the first frame read.
See also: `get_frames`
"""
if not self.has_frame(frame_idx):
raise IndexError(f"Frame index {frame_idx} out of range.")
img = self._read_frame(frame_idx)
if self.grayscale is None:
self.detect_grayscale(img)
if self.grayscale:
img = img[..., [0]]
return img
get_frames(frame_inds)
¶
Read a list of frames from the video.
Depending on the backend implementation, this may be faster than reading frames
individually using get_frame.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame_inds
|
list[int]
|
List of frame indices to read. |
required |
Returns:
| Type | Description |
|---|---|
ndarray
|
Frames as a numpy array of shape |
Notes
If the grayscale attribute is set to True, the channels dimension will
be reduced to 1 if an RGB frame is loaded from the backend.
If the grayscale attribute is set to None, the grayscale attribute
will be automatically set based on the first frame read.
See also: get_frame
Source code in sleap_io/io/video_reading.py
def get_frames(self, frame_inds: list[int]) -> np.ndarray:
"""Read a list of frames from the video.
Depending on the backend implementation, this may be faster than reading frames
individually using `get_frame`.
Args:
frame_inds: List of frame indices to read.
Returns:
Frames as a numpy array of shape `(frames, height, width, channels)` where
`channels` dimension is 1 for grayscale videos and 3 for color videos.
Notes:
If the `grayscale` attribute is set to `True`, the `channels` dimension will
be reduced to 1 if an RGB frame is loaded from the backend.
If the `grayscale` attribute is set to `None`, the `grayscale` attribute
will be automatically set based on the first frame read.
See also: `get_frame`
"""
imgs = self._read_frames(frame_inds)
if self.grayscale is None:
self.detect_grayscale(imgs[0])
if self.grayscale:
imgs = imgs[..., [0]]
return imgs
has_frame(frame_idx)
¶
Check if a frame index is contained in the video.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame_idx
|
int
|
Index of frame to check. |
required |
Returns:
| Type | Description |
|---|---|
bool
|
|
read_test_frame()
¶
Read a single frame from the video to test for grayscale.
Note
This reads the frame at index 0. This may not be appropriate if the first frame is not available in a given backend.
crop_frame(frame, crop, fill=0)
¶
Crop a frame to the specified region.
If the crop region extends beyond the frame bounds, the out-of-bounds area is filled with the fill value.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
frame
|
ndarray
|
Input frame as numpy array with shape (H, W) or (H, W, C). |
required |
crop
|
tuple[int, int, int, int]
|
Crop region as (x1, y1, x2, y2) pixel coordinates. |
required |
fill
|
tuple[int, ...] | int
|
Fill value for out-of-bounds regions. |
0
|
Returns:
| Type | Description |
|---|---|
ndarray
|
Cropped frame as numpy array. |
Source code in sleap_io/transform/frame.py
def crop_frame(
frame: np.ndarray,
crop: tuple[int, int, int, int],
fill: tuple[int, ...] | int = 0,
) -> np.ndarray:
"""Crop a frame to the specified region.
If the crop region extends beyond the frame bounds, the out-of-bounds area
is filled with the fill value.
Args:
frame: Input frame as numpy array with shape (H, W) or (H, W, C).
crop: Crop region as (x1, y1, x2, y2) pixel coordinates.
fill: Fill value for out-of-bounds regions.
Returns:
Cropped frame as numpy array.
"""
x1, y1, x2, y2 = crop
h, w = frame.shape[:2]
crop_w, crop_h = x2 - x1, y2 - y1
# Compute valid source region. Clamp the upper bounds to the lower bounds so a
# crop that lies wholly beyond the frame on an axis yields an empty (not
# negative-extent) source slice, which pastes cleanly into an all-fill output
# instead of raising a broadcast error.
src_x1 = max(0, x1)
src_y1 = max(0, y1)
src_x2 = max(src_x1, min(w, x2))
src_y2 = max(src_y1, min(h, y2))
# Extract source region
cropped = frame[src_y1:src_y2, src_x1:src_x2]
# Check if padding is needed
if x1 < 0 or y1 < 0 or x2 > w or y2 > h:
# Create output array with fill value
if frame.ndim == 3:
output_shape = (crop_h, crop_w, frame.shape[2])
else:
output_shape = (crop_h, crop_w)
output = np.full(output_shape, fill, dtype=frame.dtype)
# Compute paste region
paste_x1 = src_x1 - x1
paste_y1 = src_y1 - y1
paste_x2 = paste_x1 + (src_x2 - src_x1)
paste_y2 = paste_y1 + (src_y2 - src_y1)
output[paste_y1:paste_y2, paste_x1:paste_x2] = cropped
return output
return cropped
crop_points(points, crop)
¶
Adjust point coordinates for a crop transformation.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
points
|
ndarray
|
Coordinate array of shape (..., 2) where the last dimension contains (x, y) coordinates. NaN values are preserved. |
required |
crop
|
tuple[int, int, int, int]
|
Crop region as (x1, y1, x2, y2) pixel coordinates. |
required |
Returns:
| Type | Description |
|---|---|
ndarray
|
Adjusted coordinates with same shape as input. |
Source code in sleap_io/transform/points.py
def crop_points(
points: np.ndarray,
crop: tuple[int, int, int, int],
) -> np.ndarray:
"""Adjust point coordinates for a crop transformation.
Args:
points: Coordinate array of shape (..., 2) where the last dimension
contains (x, y) coordinates. NaN values are preserved.
crop: Crop region as (x1, y1, x2, y2) pixel coordinates.
Returns:
Adjusted coordinates with same shape as input.
"""
x1, y1, x2, y2 = crop
result = points.copy()
result[..., 0] = points[..., 0] - x1
result[..., 1] = points[..., 1] - y1
return result
get_available_image_backends()
¶
Get list of available image backend plugins.
Returns:
| Type | Description |
|---|---|
list[str]
|
List of plugin names that are currently available. Will always include "imageio" (core dependency), and may include "opencv" if installed. |
Examples:
>>> import sleap_io as sio
>>> sio.get_available_image_backends()
['imageio']
>>> 'opencv' in sio.get_available_image_backends()
False
Source code in sleap_io/io/video_reading.py
def get_available_image_backends() -> list[str]:
"""Get list of available image backend plugins.
Returns:
List of plugin names that are currently available. Will always include
"imageio" (core dependency), and may include "opencv" if installed.
Examples:
>>> import sleap_io as sio
>>> sio.get_available_image_backends()
['imageio']
>>> 'opencv' in sio.get_available_image_backends()
False
"""
return [k for k, v in _AVAILABLE_IMAGE_BACKENDS.items() if v]
get_available_video_backends()
¶
Get list of available video backend plugins.
Returns:
| Type | Description |
|---|---|
list[str]
|
List of plugin names that are currently available. Possible values include "opencv", "FFMPEG", and "pyav". |
Examples:
>>> import sleap_io as sio
>>> sio.get_available_video_backends()
['FFMPEG', 'pyav']
>>> 'opencv' in sio.get_available_video_backends()
False
Source code in sleap_io/io/video_reading.py
def get_available_video_backends() -> list[str]:
"""Get list of available video backend plugins.
Returns:
List of plugin names that are currently available. Possible values include
"opencv", "FFMPEG", and "pyav".
Examples:
>>> import sleap_io as sio
>>> sio.get_available_video_backends()
['FFMPEG', 'pyav']
>>> 'opencv' in sio.get_available_video_backends()
False
"""
return [k for k, v in _AVAILABLE_VIDEO_BACKENDS.items() if v]
get_default_image_plugin()
¶
Get the current default image plugin.
Returns:
| Type | Description |
|---|---|
str | None
|
The current default image plugin name ("opencv" or "imageio"), or None. |
Examples:
>>> import sleap_io as sio
>>> sio.get_default_image_plugin()
None
>>> sio.set_default_image_plugin("opencv")
>>> sio.get_default_image_plugin()
'opencv'
Source code in sleap_io/io/video_reading.py
def get_default_image_plugin() -> str | None:
"""Get the current default image plugin.
Returns:
The current default image plugin name ("opencv" or "imageio"), or None.
Examples:
>>> import sleap_io as sio
>>> sio.get_default_image_plugin()
None
>>> sio.set_default_image_plugin("opencv")
>>> sio.get_default_image_plugin()
'opencv'
"""
return _default_image_plugin
get_default_video_plugin()
¶
Get the current default video plugin.
Returns:
| Type | Description |
|---|---|
str | None
|
The current default video plugin name, or None if not set. |
Examples:
>>> import sleap_io as sio
>>> sio.get_default_video_plugin()
None
>>> sio.set_default_video_plugin("opencv")
>>> sio.get_default_video_plugin()
'opencv'
Source code in sleap_io/io/video_reading.py
def get_default_video_plugin() -> str | None:
"""Get the current default video plugin.
Returns:
The current default video plugin name, or None if not set.
Examples:
>>> import sleap_io as sio
>>> sio.get_default_video_plugin()
None
>>> sio.set_default_video_plugin("opencv")
>>> sio.get_default_video_plugin()
'opencv'
"""
return _default_video_plugin
get_installation_instructions(plugin=None, backend_type='video')
¶
Get installation instructions for backend plugins.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
plugin
|
str | None
|
Specific plugin name (e.g., "opencv", "FFMPEG", "pyav"), or None to get instructions for all plugins. Case-insensitive, accepts aliases. |
None
|
backend_type
|
str
|
Either "video" or "image". Determines which backend type to provide instructions for. |
'video'
|
Returns:
| Type | Description |
|---|---|
str
|
Installation instructions as a formatted string. |
Examples:
>>> import sleap_io as sio
>>> print(sio.get_installation_instructions("opencv"))
pip install sleap-io[opencv]
>>> print(sio.get_installation_instructions())
Video backend installation options:
FFMPEG (bundled): Included by default
opencv (fastest): pip install sleap-io[opencv]
pyav (balanced): pip install sleap-io[pyav]
Source code in sleap_io/io/video_reading.py
def get_installation_instructions(
plugin: str | None = None, backend_type: str = "video"
) -> str:
"""Get installation instructions for backend plugins.
Args:
plugin: Specific plugin name (e.g., "opencv", "FFMPEG", "pyav"), or None to
get instructions for all plugins. Case-insensitive, accepts aliases.
backend_type: Either "video" or "image". Determines which backend type to
provide instructions for.
Returns:
Installation instructions as a formatted string.
Examples:
>>> import sleap_io as sio
>>> print(sio.get_installation_instructions("opencv"))
pip install sleap-io[opencv]
>>> print(sio.get_installation_instructions())
Video backend installation options:
FFMPEG (bundled): Included by default
opencv (fastest): pip install sleap-io[opencv]
pyav (balanced): pip install sleap-io[pyav]
"""
if backend_type == "video":
instructions = {
"opencv": "pip install sleap-io[opencv]",
"FFMPEG": "Included by default (imageio-ffmpeg)",
"pyav": "pip install sleap-io[pyav]",
}
if plugin is not None:
plugin = normalize_plugin_name(plugin)
return instructions.get(plugin, "pip install sleap-io[all]")
else:
return (
"Video backend installation options:\n"
" FFMPEG (bundled): Included by default\n"
" opencv (fastest): pip install sleap-io[opencv]\n"
" pyav (balanced): pip install sleap-io[pyav]"
)
else:
instructions = {
"opencv": "pip install sleap-io[opencv]",
"imageio": "Already installed (core dependency)",
}
if plugin is not None:
plugin = normalize_image_plugin_name(plugin)
return instructions.get(plugin, "pip install sleap-io[all]")
else:
return (
"Image backend installation options:\n"
" opencv: pip install sleap-io[opencv]\n"
" imageio: Already installed (core dependency)"
)
normalize_image_plugin_name(plugin)
¶
Normalize image plugin names to standard format.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
plugin
|
str
|
Plugin name or alias (case-insensitive). |
required |
Returns:
| Type | Description |
|---|---|
str
|
Normalized plugin name ("opencv" or "imageio"). |
Raises:
| Type | Description |
|---|---|
ValueError
|
If plugin name is not recognized. |
Source code in sleap_io/io/video_reading.py
def normalize_image_plugin_name(plugin: str) -> str:
"""Normalize image plugin names to standard format.
Args:
plugin: Plugin name or alias (case-insensitive).
Returns:
Normalized plugin name ("opencv" or "imageio").
Raises:
ValueError: If plugin name is not recognized.
"""
plugin_lower = plugin.lower()
# Map aliases to standard names (only opencv and imageio for images)
aliases = {
"cv": "opencv",
"cv2": "opencv",
"opencv": "opencv",
"ocv": "opencv",
"imageio": "imageio",
"iio": "imageio",
}
if plugin_lower not in aliases:
raise ValueError(
f"Unknown image plugin: {plugin}. Valid options: opencv, imageio"
)
return aliases[plugin_lower]
normalize_plugin_name(plugin)
¶
Normalize plugin names to standard format.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
plugin
|
str
|
Plugin name or alias (case-insensitive). |
required |
Returns:
| Type | Description |
|---|---|
str
|
Normalized plugin name ("opencv", "FFMPEG", or "pyav"). |
Raises:
| Type | Description |
|---|---|
ValueError
|
If plugin name is not recognized. |
Source code in sleap_io/io/video_reading.py
def normalize_plugin_name(plugin: str) -> str:
"""Normalize plugin names to standard format.
Args:
plugin: Plugin name or alias (case-insensitive).
Returns:
Normalized plugin name ("opencv", "FFMPEG", or "pyav").
Raises:
ValueError: If plugin name is not recognized.
"""
plugin_lower = plugin.lower()
# Map aliases to standard names
aliases = {
"cv": "opencv",
"cv2": "opencv",
"opencv": "opencv",
"ocv": "opencv",
"ffmpeg": "FFMPEG",
"imageio-ffmpeg": "FFMPEG",
"imageio_ffmpeg": "FFMPEG",
"pyav": "pyav",
"av": "pyav",
}
if plugin_lower not in aliases:
raise ValueError(
f"Unknown plugin: {plugin}. Valid options: opencv, FFMPEG, pyav"
)
return aliases[plugin_lower]
set_default_image_plugin(plugin)
¶
Set the default image plugin for encoding/decoding embedded images.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
plugin
|
str | None
|
Image plugin name. One of "opencv" or "imageio". Also accepts aliases: "cv", "cv2", "ocv" for opencv; "iio" for imageio. Case-insensitive. If None, clears the default preference. |
required |
Examples:
>>> import sleap_io as sio
>>> sio.set_default_image_plugin("opencv")
>>> sio.set_default_image_plugin("imageio")
>>> sio.set_default_image_plugin(None) # Clear preference
Source code in sleap_io/io/video_reading.py
def set_default_image_plugin(plugin: str | None) -> None:
"""Set the default image plugin for encoding/decoding embedded images.
Args:
plugin: Image plugin name. One of "opencv" or "imageio".
Also accepts aliases: "cv", "cv2", "ocv" for opencv;
"iio" for imageio. Case-insensitive.
If None, clears the default preference.
Examples:
>>> import sleap_io as sio
>>> sio.set_default_image_plugin("opencv")
>>> sio.set_default_image_plugin("imageio")
>>> sio.set_default_image_plugin(None) # Clear preference
"""
global _default_image_plugin
if plugin is not None:
plugin = normalize_image_plugin_name(plugin)
_default_image_plugin = plugin
set_default_video_plugin(plugin)
¶
Set the default video plugin for all subsequently loaded videos.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
plugin
|
str | None
|
Video plugin name. One of "opencv", "FFMPEG", or "pyav". Also accepts aliases: "cv", "cv2", "ocv" for opencv; "imageio-ffmpeg", "imageio_ffmpeg" for FFMPEG; "av" for pyav. Case-insensitive. If None, clears the default preference. |
required |
Examples:
>>> import sleap_io as sio
>>> sio.set_default_video_plugin("opencv")
>>> sio.set_default_video_plugin("cv2") # Same as "opencv"
>>> sio.set_default_video_plugin(None) # Clear preference
Source code in sleap_io/io/video_reading.py
def set_default_video_plugin(plugin: str | None) -> None:
"""Set the default video plugin for all subsequently loaded videos.
Args:
plugin: Video plugin name. One of "opencv", "FFMPEG", or "pyav".
Also accepts aliases: "cv", "cv2", "ocv" for opencv;
"imageio-ffmpeg", "imageio_ffmpeg" for FFMPEG; "av" for pyav.
Case-insensitive. If None, clears the default preference.
Examples:
>>> import sleap_io as sio
>>> sio.set_default_video_plugin("opencv")
>>> sio.set_default_video_plugin("cv2") # Same as "opencv"
>>> sio.set_default_video_plugin(None) # Clear preference
"""
global _default_video_plugin
if plugin is not None:
plugin = normalize_plugin_name(plugin)
_default_video_plugin = plugin
uncrop_points(points, crop)
¶
Map crop-local point coordinates back to source coordinates.
Inverse of :func:crop_points: maps crop-local (x, y) coordinates back to
source coordinates by adding the crop origin (x1, y1).
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
points
|
ndarray
|
Coordinate array of shape (..., 2) where the last dimension contains (x, y) coordinates. NaN values are preserved. |
required |
crop
|
tuple[int, int, int, int]
|
Crop region as (x1, y1, x2, y2) pixel coordinates. |
required |
Returns:
| Type | Description |
|---|---|
ndarray
|
Adjusted coordinates with same shape as input. |
Source code in sleap_io/transform/points.py
def uncrop_points(
points: np.ndarray,
crop: tuple[int, int, int, int],
) -> np.ndarray:
"""Map crop-local point coordinates back to source coordinates.
Inverse of :func:`crop_points`: maps crop-local (x, y) coordinates back to
source coordinates by adding the crop origin (x1, y1).
Args:
points: Coordinate array of shape (..., 2) where the last dimension
contains (x, y) coordinates. NaN values are preserved.
crop: Crop region as (x1, y1, x2, y2) pixel coordinates.
Returns:
Adjusted coordinates with same shape as input.
"""
x1, y1, x2, y2 = crop
result = points.copy()
result[..., 0] = points[..., 0] + x1
result[..., 1] = points[..., 1] + y1
return result