pose_format.torch.masked package

pose_format.torch.masked subpackages:

• pose_format.torch.masked.tensor module
  • MaskedTensor
    • MaskedTensor.arithmetic()
    • MaskedTensor.cuda()
    • MaskedTensor.div()
    • MaskedTensor.fix_nan()
    • MaskedTensor.matmul()
    • MaskedTensor.permute()
    • MaskedTensor.pow_()
    • MaskedTensor.rename()
    • MaskedTensor.reshape()
    • MaskedTensor.size()
    • MaskedTensor.split()
    • MaskedTensor.squeeze()
    • MaskedTensor.sum()
    • MaskedTensor.to()
    • MaskedTensor.transpose()
    • MaskedTensor.zero_filled()
• pose_format.torch.masked.torch module
  • MaskedTorch
    • MaskedTorch.cat()
    • MaskedTorch.squeeze()
    • MaskedTorch.stack()
    • MaskedTorch.zeros()
  • TorchFallback
    • TorchFallback.doesnt_change_mask
• pose_format.torch.masked.torch_test module
  • TestMaskedTorch
    • TestMaskedTorch.test_cat()
    • TestMaskedTorch.test_not_implemented_method()
    • TestMaskedTorch.test_stack()
    • TestMaskedTorch.test_zeros_mask_value()
    • TestMaskedTorch.test_zeros_tensor_shape()
    • TestMaskedTorch.test_zeros_tensor_type_bool()
    • TestMaskedTorch.test_zeros_tensor_type_float()
    • TestMaskedTorch.test_zeros_tensor_value()

pose_format.torch.masked.tensor module

class pose_format.torch.masked.tensor.MaskedTensor(tensor, mask=None)

Bases: object

Container for a PyTorch tensor, providing utility functions for tensor masking.

Parameters:

• tensor (torch.Tensor) – Tensor data.

• mask (torch.Tensor, optional) – A boolean mask tensor of the same shape as tensor. If specified, elements of tensor corresponding to True values in the mask are considered valid. Defaults to a tensor of all True values.

Methods:

arithmetic(action, other)	Helper method to perform arithmetic operations on tensors.
cuda([device, non_blocking])	Moves tensor to the GPU.
div(other[, in_place, update_mask])	Performs element-wise division with another tensor.
fix_nan()	Replaces any NaN values in the tensor with zeros.
matmul(matrix)	Perform matrix multiplication.
permute(dims)	Permute dimensions of tensor.
pow_(exponent)	Raises tensor to power of a given exponent in-place.
rename(*names)	Rename tensor's dimensions.
reshape(shape)	Reshape tensor to given shape.
size(*args)	Get size of tensor for specified dimensions.
split(split_size_or_sections[, dim])	Split tensor into multiple tensors.
squeeze(dim)	Squeeze tensor along chosen dimension.
sum(dim)	Sums along a specified dimension.
to(device)	Moves tensor to a custom device.
transpose(dim0, dim1)	Transposes tensor along two dimensions.
zero_filled()	Get tensor with masked values set to zero.

arithmetic(action, other)

Helper method to perform arithmetic operations on tensors.

Parameters:

• action (str) – The arithmetic operation to be performed.

• other (Union[MaskedTensor`, torch.Tensor, float, int]) – The second operand.

Returns:

New MaskedTensor after the operation.

Return type:

MaskedTensor

cuda(device=None, non_blocking=False)

Moves tensor to the GPU.

Parameters:

• device (str or torch.device, optional) – The target CUDA device.

• non_blocking (bool, optional) – Whether to perform an operation asynchronously. Default is False.

Returns:

Tensor on CUDA device.

Return type:

MaskedTensor

div(other, in_place=False, update_mask=True)

Performs element-wise division with another tensor.

Parameters:

• other (MaskedTensor) – The tensor to divide with.

• in_place (bool, optional) – If True, performs the operation in-place. Default is False.

• update_mask (bool, optional) – If True, updates the mask after division. Default is True.

Returns:

Resultant tensor after division.

Return type:

MaskedTensor

fix_nan()

Replaces any NaN values in the tensor with zeros.

Returns:

Tensor with NaN values replaced by zeros.

Return type:

MaskedTensor

matmul(matrix)

Perform matrix multiplication.

Parameters:

matrix (torch.Tensor) – matrix to multiply with.

Returns:

New masked tensor after multiplication.

Return type:

MaskedTensor

permute(dims)

Permute dimensions of tensor.

Parameters:

dims (tuple) – Desired ordering of dimensions.

Returns:

Permuted masked tensor.

Return type:

MaskedTensor

pow_(exponent)

Raises tensor to power of a given exponent in-place.

Parameters:

exponent (float) – The exponent value.

Returns:

Masked tensor raised to a given exponent.

Return type:

MaskedTensor

rename(*names)

Rename tensor’s dimensions.

Parameters:

names (tuple) – Desired names for each dimension.

Returns:

Renamed masked tensor.

Return type:

MaskedTensor

reshape(shape)

Reshape tensor to given shape.

Parameters:

shape (tuple) – Desired shape.

Returns:

Reshaped tensor.

Return type:

MaskedTensor

size(*args)

Get size of tensor for specified dimensions.

Returns:

Size of tensor.

Return type:

torch.Size

split(split_size_or_sections, dim=0)

Split tensor into multiple tensors.

Parameters:

• split_size_or_sections (int or tuple) – Size or sections to split tensor.

• dim (int, optional) – Dimension along which to split tensor. Default is 0.

Returns:

List of split tensors.

Return type:

list[MaskedTensor]

squeeze(dim)

Squeeze tensor along chosen dimension.

Parameters:

dim (int) – Dimension to squeeze.

Returns:

Squeezed masked tensor.

Return type:

MaskedTensor

sum(dim)

Sums along a specified dimension.

Parameters:

dim (int) – dimension to sum over.

Returns:

Summed tensor along the specified dimension.

Return type:

MaskedTensor

to(device)

Moves tensor to a custom device.

Parameters:

device (str or torch.device) – The target device.

Returns:

Tensor on the other device.

Return type:

MaskedTensor

transpose(dim0, dim1)

Transposes tensor along two dimensions.

Parameters:

• dim0 (int) – Two dimensions to which to transpose.

• dim1 (int) – Two dimensions to which to transpose.

Returns:

Transposed masked tensor.

Return type:

MaskedTensor

zero_filled()

Get tensor with masked values set to zero.

Returns:

Tensor with masked values set to zero.

Return type:

torch.Tensor

pose_format.torch.masked.torch module

Classes:

MaskedTorch	class mimicing torch functions and giving support for MaskedTensor.
TorchFallback	Meta class that gives a fallback mechanism to use torch functions on MaskedTensor objects.

class pose_format.torch.masked.torch.MaskedTorch

Bases: object

class mimicing torch functions and giving support for MaskedTensor.

Methods:

cat(tensors, dim)	Concatenate MaskedTensor objects along a specified dimension.
squeeze(masked_tensor)	Remove dimensions of size 1 from MaskedTensor.
stack(tensors, dim)	Stack MaskedTensor objects along a new dimension.
zeros(*size[, dtype])	Creates a MaskedTensor of zeros with a given shape and data type.

static cat(tensors, dim)

Concatenate MaskedTensor objects along a specified dimension.

Parameters:

• tensors (list) – List of tensors or MaskedTensor objects to be concatenated.

• dim (int) – Dimension along to concatenate.

Returns:

Concatenated tensor.

Return type:

MaskedTensor

static squeeze(masked_tensor)

Remove dimensions of size 1 from MaskedTensor.

Parameters:

masked_tensor (MaskedTensor) – tensor from which dimensions are to be removed.

Returns:

Squeezed masked tensor.

Return type:

MaskedTensor

static stack(tensors, dim)

Stack MaskedTensor objects along a new dimension.

Parameters:

• tensors (list) – List of MaskedTensor objects to be stacked.

• dim (int) – New dimension along which to stack.

Returns:

Stacked maked tensor.

Return type:

MaskedTensor

static zeros(*size, dtype=None)

Creates a MaskedTensor of zeros with a given shape and data type.

Parameters:

• *size (ints) – Dimensions of desired tensor.

• dtype (torch.dtype, optional) – Data type of the tensor. If None, defaults to torch.float.

Returns:

masked tensor filled with zeros.

Return type:

MaskedTensor

class pose_format.torch.masked.torch.TorchFallback

Bases: type

Meta class that gives a fallback mechanism to use torch functions on MaskedTensor objects. :noindex:

Attributes:

doesnt_change_mask

doesnt_change_mask = {'acos', 'asin', 'atan', 'cos', 'sin', 'sqrt', 'square', 'tan', 'unsqueeze'}

pose_format.torch.masked.torch_test module

Classes:

TestMaskedTorch([methodName])

Test cases for the MaskedTensor class

class pose_format.torch.masked.torch_test.TestMaskedTorch(methodName='runTest')

Bases: TestCase

Test cases for the MaskedTensor class

Methods:

test_cat()	Test cat method for concatenating MaskedTensor objects along a specified dimension.
test_not_implemented_method()	Tests behavior when invoking an unimplemented method on a MaskedTensor.
test_stack()	Tests stack method for stacking MaskedTensor objects along a new dimension.
test_zeros_mask_value()	Test if the mask in the produced zeros MaskedTensor is initialized with zero values.
test_zeros_tensor_shape()	Test if zeros method correctly produces a MaskedTensor with the desired shape.
test_zeros_tensor_type_bool()	Test if the zeros method produces a MaskedTensor with the correct boolean data type.
test_zeros_tensor_type_float()	Test if the zeros method produces a MaskedTensor with the correct float data type.
test_zeros_tensor_value()	Test if the zeros method produces a MaskedTensor with all zero values.

test_cat()

Test cat method for concatenating MaskedTensor objects along a specified dimension.

test_not_implemented_method()

Tests behavior when invoking an unimplemented method on a MaskedTensor.

test_stack()

Tests stack method for stacking MaskedTensor objects along a new dimension.

test_zeros_mask_value()

Test if the mask in the produced zeros MaskedTensor is initialized with zero values.

test_zeros_tensor_shape()

Test if zeros method correctly produces a MaskedTensor with the desired shape.

test_zeros_tensor_type_bool()

Test if the zeros method produces a MaskedTensor with the correct boolean data type.

test_zeros_tensor_type_float()

Test if the zeros method produces a MaskedTensor with the correct float data type.

test_zeros_tensor_value()

Test if the zeros method produces a MaskedTensor with all zero values.