pose_format.pose_body module

Classes:

PoseBody(fps, data, confidence)

Class for body data of a pose.

class pose_format.pose_body.PoseBody(fps, data, confidence)[source]

Bases: object

Class for body data of a pose.

Parameters:

fps (float) – Frames per second.
data – Data in the format (Frames, People, Points, Dims) e.g., (93, 1, 137, 2).
confidence – Confidence data in the format (Frames, People, Points) e.g., (93, 1, 137).

Methods:

`augment2d`([rotation_std, shear_std, scale_std])	Augment 2D data with given standard deviations.
`bbox`(header)	For computing bounding box of PoseBody.
`flatten`()	Converts data from the (Frames, People, Points, Dims) masked representation to an array of points.
`frame_dropout_given_percent`(dropout_percent)	Drop of frames based on given dropout percentage.
`frame_dropout_normal`([dropout_mean, dropout_std])	drop frames depending on normal distribution with given mean and standard deviation.
`frame_dropout_uniform`([dropout_min, dropout_max])	Randomly drops frames depending on a uniform distribution - given minimum and maximum percentages.
`get_points`(indexes)	Get points from PoseBody.
`matmul`(matrix)	Multiplies PoseBody data with a numpy.ndarray matrix.
`numpy`()	Convert the current PoseBody representation to NumpyPoseBody.
`points_perspective`()	Give points in PoseBody as a perspective view.
`read`(header, reader, **kwargs)	Reads pose data a buffer (BufferReader) based on the header's version.
`read_v0_0`(header, reader, **unused_kwargs)	reads version 0.0 pose data.
`read_v0_1`(header, reader[, start_frame, ...])	Reads pose data for version 0.1 from a buffer.
`read_v0_1_frames`(frames, shape, reader[, ...])	Reads frame data for version 0.1 from a buffer.
`select_frames`(frame_indexes)	Selects specific frames from PoseBody object.
`slice_step`(by)	Slices data by skipping rows.
`tensorflow`()	Converts current PoseBody representation to TensorflowPoseBody.
`torch`()	Converts current PoseBody to TorchPoseBody.
`write`(version, buffer)	Writes data to a file based on version of spec: in docs/spec.
`zero_filled`()	Creates a new PoseBody instance with data replaced by zeros.

Attributes:

tensor_reader

augment2d(rotation_std=0.2, shear_std=0.2, scale_std=0.2)[source]

Augment 2D data with given standard deviations.

Parameters:

rotation_std (float, optional) – Rotation in radians. Default is 0.2.
shear_std (float, optional) – Shear X in percent. Default is 0.2.
scale_std (float, optional) – Scale X in percent. Default is 0.2.

Returns:

Augmented PoseBody instance.

Return type:

PoseBody

Note

The method modifies the PoseBody based on shear, rotation, and scaling.
shear_std based on https://en.wikipedia.org/wiki/Shear_matrix
rotation_std based on https://en.wikipedia.org/wiki/Rotation_matrix
scale_std based on https://en.wikipedia.org/wiki/Scaling_(geometry)

bbox(header)[source]

For computing bounding box of PoseBody.

Parameters:: header (PoseHeader) – Header containing the version of the pose data.
Returns:: PoseBody instance with bounding box.
Return type:: PoseBody
Raises:: NotImplementedError – If the bbox is not implemented in class.

flatten()[source]

Converts data from the (Frames, People, Points, Dims) masked representation to an array of points.

Every item in the result array contains the following dimensions: 0. Time in milliseconds 1. Person ID 2. Point ID 3. X dimension 4. Y dimension 5. Z dimension (if exists) 6. Pose estimation confidence

Returns:: Array of points with detailed dimensions.
Return type:: np.ndarray
Raises:: NotImplementedError – If the method is not implemented for the specific class.

frame_dropout_given_percent(dropout_percent)[source]

Drop of frames based on given dropout percentage.

Parameters:

dropout_percent (float) – Percentage of frames to drop. Between 0 and 1 (e.g., 0.2 means drop 20% of the frames).

Returns:

New PoseBody object with the gotten frames.
List of frame indexes.

Return type:

Tuple[PoseBody, List[int]]

Note

Actual number of dropped frames might be slightly different due to rounding!

frame_dropout_normal(dropout_mean=0.5, dropout_std=0.1)[source]

drop frames depending on normal distribution with given mean and standard deviation.

Parameters:

dropout_mean (float, optional) – Mean percentage of frames to drop. Default is 0.5.
dropout_std (float, optional) – Standard deviation of percentage of frames to drop. Default is 0.1.

Returns:

New PoseBody object with dropped frames.
List of retrieved frame indexes.

Return type:

Tuple[PoseBody, List[int]]

frame_dropout_uniform(dropout_min=0.2, dropout_max=1.0)[source]

Randomly drops frames depending on a uniform distribution - given minimum and maximum percentages.

Parameters:

dropout_min (float, optional) – Minimum percentage of frames to drop. Default is 0.2.
dropout_max (float, optional) – Maximum percentage of frames to drop. Default is 1.0.

Returns:

New PoseBody object with dropped frames.
List of frame indexes that were retained.

Return type:

Tuple[PoseBody, List[int]]

get_points(indexes)[source]

Get points from PoseBody.

Parameters:: indexes (List[int]) – List of point indices to get from PoseBody.
Returns:: PoseBody instance containing only choosen points.
Return type:: PoseBody
Raises:: NotImplementedError – If the get_points is not implemented in class.

matmul(matrix)[source]

Multiplies PoseBody data with a numpy.ndarray matrix.

Parameters:: matrix (np.ndarray) – The matrix to multiply the PoseBody data with.
Returns:: PoseBody instance with data multiplied by a numpy array.
Return type:: PoseBody
Raises:: NotImplementedError – If the matmul is not implemented in class.

numpy()[source]

Convert the current PoseBody representation to NumpyPoseBody.

Returns:: The converted PoseBody object.
Return type:: NumpyPoseBody
Raises:: NotImplementedError – If numpy is not implemented.

points_perspective()[source]

Give points in PoseBody as a perspective view.

Returns:: PoseBody instance with points adjusted for perspective.
Return type:: PoseBody
Raises:: NotImplementedError – If the method is not implemented for the specific class.

classmethod read(header, reader, **kwargs)[source]

Reads pose data a buffer (BufferReader) based on the header’s version.

Parameters:

header (PoseHeader) – Header containing the version of its pose data.
reader (BufferReader) – Buffer from which to read the pose data.
**kwargs (dict) – Additional parameters for reading specific versions.

Returns:

PoseBody object initialized with the read data.

Return type:

PoseBody

Raises:

NotImplementedError – If header’s version is not supported / unknown.

classmethod read_v0_0(header, reader, **unused_kwargs)[source]

reads version 0.0 pose data.

Parameters:

header (PoseHeader) – Header containing the version of the pose data.
reader (BufferReader) – Buffer from which to read the pose data.
unused_kwargs (dict) – Unused additional parameters for this version.

Raises:

NotImplementedError – method for this version is not implemented.

classmethod read_v0_1(header, reader, start_frame=None, end_frame=None, **unused_kwargs)[source]

Reads pose data for version 0.1 from a buffer.

Parameters:

header (PoseHeader) – Header containing the version of the pose data.
reader (BufferReader) – Buffer from which to read the pose data.
start_frame (int, optional) – Index of the first frame to read. Default is None.
end_frame (int, optional) – Index of the last frame to read. Default is None.
**unused_kwargs (dict) – Unused additional parameters for this version.

Returns:

PoseBody object initialized with the read data for version 0.1.

Return type:

PoseBody

classmethod read_v0_1_frames(frames, shape, reader, start_frame=None, end_frame=None)[source]

Reads frame data for version 0.1 from a buffer.

Parameters:

frames (int) – Number of frames in the pose data.
shape (List[int]) – Shape of the pose data.
reader (BufferReader) – Buffer from which to read the pose data.
start_frame (int, optional) – Index of the first frame to read. Default is None.
end_frame (int, optional) – Index of the last frame to read. Default is None.

Returns:

Array containing the pose data for the specified frames.

Return type:

ndarray

Raises:

ValueError – If start_frame is greater than number of frames.

select_frames(frame_indexes)[source]

Selects specific frames from PoseBody object.

Parameters:: frame_indexes (List[int]) – List of frame indexes to select.
Returns:: PoseBody object containing only the selected frames.
Return type:: PoseBody
Raises:: IndexError – If any of the specified frame indices are out of the valid range for the current PoseBody data.

slice_step(by)[source]

Slices data by skipping rows. This affects the fps (frames per seconds).

Parameters:: by (int) – Take one row every “by” rows.
Returns:: PoseBody instance with sliced data.
Return type:: PoseBody

tensor_reader = 'ABSTRACT-DO-NOT-USE'

tensorflow()[source]

Converts current PoseBody representation to TensorflowPoseBody.

Returns:: Converted PoseBody object.
Return type:: TensorflowPoseBody
Raises:: NotImplementedError – If tensorflow is not implemented.

torch()[source]

Converts current PoseBody to TorchPoseBody.

Returns:: The converted PoseBody object.
Return type:: TorchPoseBody
Raises:: NotImplementedError – If toch is not implemented.

write(version, buffer)[source]

Writes data to a file based on version of spec: in docs/spec.

Parameters:

version (float) – Version of the pose data to write.
buffer (BinaryIO) – Buffer to write the pose data to.

zero_filled()[source]

Creates a new PoseBody instance with data replaced by zeros.

Returns:: PoseBody instance with zero-filled data.
Return type:: PoseBody
Raises:: NotImplementedError – If the zero_filled is not implemented on class .