liegroups.torch

The PyTorch implementation uses torch.Tensor as the backend linear algebra library, which allows the user to on the GPU or CPU and integrate with other aspects of PyTorch.

This version provides sensible options for batching the transformations themselves, as well as anything they might operate on, and is generally agnostic to the specific Tensor type (e.g., given a torch.cuda.FloatTensor as input, the output will also be a torch.cuda.FloatTensor).

liegroups.torch.SO2

alias of liegroups.torch.so2.SO2Matrix

class liegroups.torch.so2.SO2Matrix(mat)

See liegroups.SO2

cpu()

Return a copy with the underlying tensor on the CPU.

cuda(device=None, non_blocking=False)

Return a copy with the underlying tensor on the GPU.

classmethod from_numpy(other, pin_memory=False)

Create a torch-based copy of a numpy-based rotation.

is_cuda()

Returns true if the underlying tensor is a CUDA tensor

is_pinned()

Returns true if the underlying tensor resides in pinned memory

pin_memory()

Return a copy with the underlying tensor in pinned (page-locked) memory. Makes host-to-GPU copies faster.

See: http://pytorch.org/docs/master/notes/cuda.html?highlight=pinned

liegroups.torch.SE2

alias of liegroups.torch.se2.SE2Matrix

class liegroups.torch.se2.SE2Matrix(rot, trans)

See liegroups.SE2

cpu()

Return a copy with the underlying tensors on the CPU.

cuda(device=None, non_blocking=False)

Return a copy with the underlying tensors on the GPU.

classmethod from_numpy(other, pin_memory=False)

Create a torch-based copy of a numpy-based transformation.

is_cuda()

Returns true if the underlying tensors are CUDA tensors

is_pinned()

Returns true if the underlying tensors reside in pinned memory

pin_memory()

Return a copy with the underlying tensor in pinned (page-locked) memory. Makes host-to-GPU copies faster.

See: http://pytorch.org/docs/master/notes/cuda.html?highlight=pinned

liegroups.torch.SO3

alias of liegroups.torch.so3.SO3Matrix

class liegroups.torch.so3.SO3Matrix(mat)

See liegroups.SO3

cpu()

Return a copy with the underlying tensor on the CPU.

cuda(device=None, non_blocking=False)

Return a copy with the underlying tensor on the GPU.

classmethod from_numpy(other, pin_memory=False)

Create a torch-based copy of a numpy-based rotation.

is_cuda()

Returns true if the underlying tensor is a CUDA tensor

is_pinned()

Returns true if the underlying tensor resides in pinned memory

pin_memory()

Return a copy with the underlying tensor in pinned (page-locked) memory. Makes host-to-GPU copies faster.

See: http://pytorch.org/docs/master/notes/cuda.html?highlight=pinned

liegroups.torch.SE3

alias of liegroups.torch.se3.SE3Matrix

class liegroups.torch.se3.SE3Matrix(rot, trans)

See liegroups.SE3

cpu()

Return a copy with the underlying tensors on the CPU.

cuda(device=None, non_blocking=False)

Return a copy with the underlying tensors on the GPU.

classmethod from_numpy(other, pin_memory=False)

Create a torch-based copy of a numpy-based transformation.

is_cuda()

Returns true if the underlying tensors are CUDA tensors

is_pinned()

Returns true if the underlying tensors reside in pinned memory

pin_memory()

Return a copy with the underlying tensor in pinned (page-locked) memory. Makes host-to-GPU copies faster.

See: http://pytorch.org/docs/master/notes/cuda.html?highlight=pinned