SMPLer-X / assets /conversions.py
onescotch
add huggingface implementation
2de1f98
raw
history blame
16.5 kB
import torch
import torch.nn as nn
import torchgeometry as tgm
__all__ = [
# functional api
"pi",
"rad2deg",
"deg2rad",
"convert_points_from_homogeneous",
"convert_points_to_homogeneous",
"angle_axis_to_rotation_matrix",
"rotation_matrix_to_angle_axis",
"rotation_matrix_to_quaternion",
"quaternion_to_angle_axis",
"angle_axis_to_quaternion",
"rtvec_to_pose",
# layer api
"RadToDeg",
"DegToRad",
"ConvertPointsFromHomogeneous",
"ConvertPointsToHomogeneous",
]
"""Constant with number pi
"""
pi = torch.Tensor([3.14159265358979323846])
def rad2deg(tensor):
r"""Function that converts angles from radians to degrees.
See :class:`~torchgeometry.RadToDeg` for details.
Args:
tensor (Tensor): Tensor of arbitrary shape.
Returns:
Tensor: Tensor with same shape as input.
Example:
>>> input = tgm.pi * torch.rand(1, 3, 3)
>>> output = tgm.rad2deg(input)
"""
if not torch.is_tensor(tensor):
raise TypeError("Input type is not a torch.Tensor. Got {}"
.format(type(tensor)))
return 180. * tensor / pi.to(tensor.device).type(tensor.dtype)
def deg2rad(tensor):
r"""Function that converts angles from degrees to radians.
See :class:`~torchgeometry.DegToRad` for details.
Args:
tensor (Tensor): Tensor of arbitrary shape.
Returns:
Tensor: Tensor with same shape as input.
Examples::
>>> input = 360. * torch.rand(1, 3, 3)
>>> output = tgm.deg2rad(input)
"""
if not torch.is_tensor(tensor):
raise TypeError("Input type is not a torch.Tensor. Got {}"
.format(type(tensor)))
return tensor * pi.to(tensor.device).type(tensor.dtype) / 180.
def convert_points_from_homogeneous(points):
r"""Function that converts points from homogeneous to Euclidean space.
See :class:`~torchgeometry.ConvertPointsFromHomogeneous` for details.
Examples::
>>> input = torch.rand(2, 4, 3) # BxNx3
>>> output = tgm.convert_points_from_homogeneous(input) # BxNx2
"""
if not torch.is_tensor(points):
raise TypeError("Input type is not a torch.Tensor. Got {}".format(
type(points)))
if len(points.shape) < 2:
raise ValueError("Input must be at least a 2D tensor. Got {}".format(
points.shape))
return points[..., :-1] / points[..., -1:]
def convert_points_to_homogeneous(points):
r"""Function that converts points from Euclidean to homogeneous space.
See :class:`~torchgeometry.ConvertPointsToHomogeneous` for details.
Examples::
>>> input = torch.rand(2, 4, 3) # BxNx3
>>> output = tgm.convert_points_to_homogeneous(input) # BxNx4
"""
if not torch.is_tensor(points):
raise TypeError("Input type is not a torch.Tensor. Got {}".format(
type(points)))
if len(points.shape) < 2:
raise ValueError("Input must be at least a 2D tensor. Got {}".format(
points.shape))
return nn.functional.pad(points, (0, 1), "constant", 1.0)
def angle_axis_to_rotation_matrix(angle_axis):
"""Convert 3d vector of axis-angle rotation to 4x4 rotation matrix
Args:
angle_axis (Tensor): tensor of 3d vector of axis-angle rotations.
Returns:
Tensor: tensor of 4x4 rotation matrices.
Shape:
- Input: :math:`(N, 3)`
- Output: :math:`(N, 4, 4)`
Example:
>>> input = torch.rand(1, 3) # Nx3
>>> output = tgm.angle_axis_to_rotation_matrix(input) # Nx4x4
"""
def _compute_rotation_matrix(angle_axis, theta2, eps=1e-6):
# We want to be careful to only evaluate the square root if the
# norm of the angle_axis vector is greater than zero. Otherwise
# we get a division by zero.
k_one = 1.0
theta = torch.sqrt(theta2)
wxyz = angle_axis / (theta + eps)
wx, wy, wz = torch.chunk(wxyz, 3, dim=1)
cos_theta = torch.cos(theta)
sin_theta = torch.sin(theta)
r00 = cos_theta + wx * wx * (k_one - cos_theta)
r10 = wz * sin_theta + wx * wy * (k_one - cos_theta)
r20 = -wy * sin_theta + wx * wz * (k_one - cos_theta)
r01 = wx * wy * (k_one - cos_theta) - wz * sin_theta
r11 = cos_theta + wy * wy * (k_one - cos_theta)
r21 = wx * sin_theta + wy * wz * (k_one - cos_theta)
r02 = wy * sin_theta + wx * wz * (k_one - cos_theta)
r12 = -wx * sin_theta + wy * wz * (k_one - cos_theta)
r22 = cos_theta + wz * wz * (k_one - cos_theta)
rotation_matrix = torch.cat(
[r00, r01, r02, r10, r11, r12, r20, r21, r22], dim=1)
return rotation_matrix.view(-1, 3, 3)
def _compute_rotation_matrix_taylor(angle_axis):
rx, ry, rz = torch.chunk(angle_axis, 3, dim=1)
k_one = torch.ones_like(rx)
rotation_matrix = torch.cat(
[k_one, -rz, ry, rz, k_one, -rx, -ry, rx, k_one], dim=1)
return rotation_matrix.view(-1, 3, 3)
# stolen from ceres/rotation.h
_angle_axis = torch.unsqueeze(angle_axis, dim=1)
theta2 = torch.matmul(_angle_axis, _angle_axis.transpose(1, 2))
theta2 = torch.squeeze(theta2, dim=1)
# compute rotation matrices
rotation_matrix_normal = _compute_rotation_matrix(angle_axis, theta2)
rotation_matrix_taylor = _compute_rotation_matrix_taylor(angle_axis)
# create mask to handle both cases
eps = 1e-6
mask = (theta2 > eps).view(-1, 1, 1).to(theta2.device)
mask_pos = (mask).type_as(theta2)
mask_neg = (mask == False).type_as(theta2) # noqa
# create output pose matrix
batch_size = angle_axis.shape[0]
rotation_matrix = torch.eye(4).to(angle_axis.device).type_as(angle_axis)
rotation_matrix = rotation_matrix.view(1, 4, 4).repeat(batch_size, 1, 1)
# fill output matrix with masked values
rotation_matrix[..., :3, :3] = \
mask_pos * rotation_matrix_normal + mask_neg * rotation_matrix_taylor
return rotation_matrix # Nx4x4
def rtvec_to_pose(rtvec):
"""
Convert axis-angle rotation and translation vector to 4x4 pose matrix
Args:
rtvec (Tensor): Rodrigues vector transformations
Returns:
Tensor: transformation matrices
Shape:
- Input: :math:`(N, 6)`
- Output: :math:`(N, 4, 4)`
Example:
>>> input = torch.rand(3, 6) # Nx6
>>> output = tgm.rtvec_to_pose(input) # Nx4x4
"""
assert rtvec.shape[-1] == 6, 'rtvec=[rx, ry, rz, tx, ty, tz]'
pose = angle_axis_to_rotation_matrix(rtvec[..., :3])
pose[..., :3, 3] = rtvec[..., 3:]
return pose
def rotation_matrix_to_angle_axis(rotation_matrix):
"""Convert 3x4 rotation matrix to Rodrigues vector
Args:
rotation_matrix (Tensor): rotation matrix.
Returns:
Tensor: Rodrigues vector transformation.
Shape:
- Input: :math:`(N, 3, 4)`
- Output: :math:`(N, 3)`
Example:
>>> input = torch.rand(2, 3, 4) # Nx4x4
>>> output = tgm.rotation_matrix_to_angle_axis(input) # Nx3
"""
# todo add check that matrix is a valid rotation matrix
quaternion = rotation_matrix_to_quaternion(rotation_matrix)
return quaternion_to_angle_axis(quaternion)
def rotation_matrix_to_quaternion(rotation_matrix, eps=1e-6):
"""Convert 3x4 rotation matrix to 4d quaternion vector
This algorithm is based on algorithm described in
https://github.com/KieranWynn/pyquaternion/blob/master/pyquaternion/quaternion.py#L201
Args:
rotation_matrix (Tensor): the rotation matrix to convert.
Return:
Tensor: the rotation in quaternion
Shape:
- Input: :math:`(N, 3, 4)`
- Output: :math:`(N, 4)`
Example:
>>> input = torch.rand(4, 3, 4) # Nx3x4
>>> output = tgm.rotation_matrix_to_quaternion(input) # Nx4
"""
if not torch.is_tensor(rotation_matrix):
raise TypeError("Input type is not a torch.Tensor. Got {}".format(
type(rotation_matrix)))
if len(rotation_matrix.shape) > 3:
raise ValueError(
"Input size must be a three dimensional tensor. Got {}".format(
rotation_matrix.shape))
if not rotation_matrix.shape[-2:] == (3, 4):
raise ValueError(
"Input size must be a N x 3 x 4 tensor. Got {}".format(
rotation_matrix.shape))
rmat_t = torch.transpose(rotation_matrix, 1, 2)
mask_d2 = rmat_t[:, 2, 2] < eps
mask_d0_d1 = rmat_t[:, 0, 0] > rmat_t[:, 1, 1]
mask_d0_nd1 = rmat_t[:, 0, 0] < -rmat_t[:, 1, 1]
t0 = 1 + rmat_t[:, 0, 0] - rmat_t[:, 1, 1] - rmat_t[:, 2, 2]
q0 = torch.stack([rmat_t[:, 1, 2] - rmat_t[:, 2, 1],
t0, rmat_t[:, 0, 1] + rmat_t[:, 1, 0],
rmat_t[:, 2, 0] + rmat_t[:, 0, 2]], -1)
t0_rep = t0.repeat(4, 1).t()
t1 = 1 - rmat_t[:, 0, 0] + rmat_t[:, 1, 1] - rmat_t[:, 2, 2]
q1 = torch.stack([rmat_t[:, 2, 0] - rmat_t[:, 0, 2],
rmat_t[:, 0, 1] + rmat_t[:, 1, 0],
t1, rmat_t[:, 1, 2] + rmat_t[:, 2, 1]], -1)
t1_rep = t1.repeat(4, 1).t()
t2 = 1 - rmat_t[:, 0, 0] - rmat_t[:, 1, 1] + rmat_t[:, 2, 2]
q2 = torch.stack([rmat_t[:, 0, 1] - rmat_t[:, 1, 0],
rmat_t[:, 2, 0] + rmat_t[:, 0, 2],
rmat_t[:, 1, 2] + rmat_t[:, 2, 1], t2], -1)
t2_rep = t2.repeat(4, 1).t()
t3 = 1 + rmat_t[:, 0, 0] + rmat_t[:, 1, 1] + rmat_t[:, 2, 2]
q3 = torch.stack([t3, rmat_t[:, 1, 2] - rmat_t[:, 2, 1],
rmat_t[:, 2, 0] - rmat_t[:, 0, 2],
rmat_t[:, 0, 1] - rmat_t[:, 1, 0]], -1)
t3_rep = t3.repeat(4, 1).t()
mask_c0 = mask_d2 * mask_d0_d1
mask_c1 = mask_d2 * ~(mask_d0_d1)
mask_c2 = ~(mask_d2) * mask_d0_nd1
mask_c3 = ~(mask_d2) * ~(mask_d0_nd1)
mask_c0 = mask_c0.view(-1, 1).type_as(q0)
mask_c1 = mask_c1.view(-1, 1).type_as(q1)
mask_c2 = mask_c2.view(-1, 1).type_as(q2)
mask_c3 = mask_c3.view(-1, 1).type_as(q3)
q = q0 * mask_c0 + q1 * mask_c1 + q2 * mask_c2 + q3 * mask_c3
q /= torch.sqrt(t0_rep * mask_c0 + t1_rep * mask_c1 + # noqa
t2_rep * mask_c2 + t3_rep * mask_c3) # noqa
q *= 0.5
return q
def quaternion_to_angle_axis(quaternion: torch.Tensor) -> torch.Tensor:
"""Convert quaternion vector to angle axis of rotation.
Adapted from ceres C++ library: ceres-solver/include/ceres/rotation.h
Args:
quaternion (torch.Tensor): tensor with quaternions.
Return:
torch.Tensor: tensor with angle axis of rotation.
Shape:
- Input: :math:`(*, 4)` where `*` means, any number of dimensions
- Output: :math:`(*, 3)`
Example:
>>> quaternion = torch.rand(2, 4) # Nx4
>>> angle_axis = tgm.quaternion_to_angle_axis(quaternion) # Nx3
"""
if not torch.is_tensor(quaternion):
raise TypeError("Input type is not a torch.Tensor. Got {}".format(
type(quaternion)))
if not quaternion.shape[-1] == 4:
raise ValueError("Input must be a tensor of shape Nx4 or 4. Got {}"
.format(quaternion.shape))
# unpack input and compute conversion
q1: torch.Tensor = quaternion[..., 1]
q2: torch.Tensor = quaternion[..., 2]
q3: torch.Tensor = quaternion[..., 3]
sin_squared_theta: torch.Tensor = q1 * q1 + q2 * q2 + q3 * q3
sin_theta: torch.Tensor = torch.sqrt(sin_squared_theta)
cos_theta: torch.Tensor = quaternion[..., 0]
two_theta: torch.Tensor = 2.0 * torch.where(
cos_theta < 0.0,
torch.atan2(-sin_theta, -cos_theta),
torch.atan2(sin_theta, cos_theta))
k_pos: torch.Tensor = two_theta / sin_theta
k_neg: torch.Tensor = 2.0 * torch.ones_like(sin_theta)
k: torch.Tensor = torch.where(sin_squared_theta > 0.0, k_pos, k_neg)
angle_axis: torch.Tensor = torch.zeros_like(quaternion)[..., :3]
angle_axis[..., 0] += q1 * k
angle_axis[..., 1] += q2 * k
angle_axis[..., 2] += q3 * k
return angle_axis
# based on:
# https://github.com/facebookresearch/QuaterNet/blob/master/common/quaternion.py#L138
def angle_axis_to_quaternion(angle_axis: torch.Tensor) -> torch.Tensor:
"""Convert an angle axis to a quaternion.
Adapted from ceres C++ library: ceres-solver/include/ceres/rotation.h
Args:
angle_axis (torch.Tensor): tensor with angle axis.
Return:
torch.Tensor: tensor with quaternion.
Shape:
- Input: :math:`(*, 3)` where `*` means, any number of dimensions
- Output: :math:`(*, 4)`
Example:
>>> angle_axis = torch.rand(2, 4) # Nx4
>>> quaternion = tgm.angle_axis_to_quaternion(angle_axis) # Nx3
"""
if not torch.is_tensor(angle_axis):
raise TypeError("Input type is not a torch.Tensor. Got {}".format(
type(angle_axis)))
if not angle_axis.shape[-1] == 3:
raise ValueError("Input must be a tensor of shape Nx3 or 3. Got {}"
.format(angle_axis.shape))
# unpack input and compute conversion
a0: torch.Tensor = angle_axis[..., 0:1]
a1: torch.Tensor = angle_axis[..., 1:2]
a2: torch.Tensor = angle_axis[..., 2:3]
theta_squared: torch.Tensor = a0 * a0 + a1 * a1 + a2 * a2
theta: torch.Tensor = torch.sqrt(theta_squared)
half_theta: torch.Tensor = theta * 0.5
mask: torch.Tensor = theta_squared > 0.0
ones: torch.Tensor = torch.ones_like(half_theta)
k_neg: torch.Tensor = 0.5 * ones
k_pos: torch.Tensor = torch.sin(half_theta) / theta
k: torch.Tensor = torch.where(mask, k_pos, k_neg)
w: torch.Tensor = torch.where(mask, torch.cos(half_theta), ones)
quaternion: torch.Tensor = torch.zeros_like(angle_axis)
quaternion[..., 0:1] += a0 * k
quaternion[..., 1:2] += a1 * k
quaternion[..., 2:3] += a2 * k
return torch.cat([w, quaternion], dim=-1)
# TODO: add below funtionalities
# - pose_to_rtvec
# layer api
class RadToDeg(nn.Module):
r"""Creates an object that converts angles from radians to degrees.
Args:
tensor (Tensor): Tensor of arbitrary shape.
Returns:
Tensor: Tensor with same shape as input.
Examples::
>>> input = tgm.pi * torch.rand(1, 3, 3)
>>> output = tgm.RadToDeg()(input)
"""
def __init__(self):
super(RadToDeg, self).__init__()
def forward(self, input):
return rad2deg(input)
class DegToRad(nn.Module):
r"""Function that converts angles from degrees to radians.
Args:
tensor (Tensor): Tensor of arbitrary shape.
Returns:
Tensor: Tensor with same shape as input.
Examples::
>>> input = 360. * torch.rand(1, 3, 3)
>>> output = tgm.DegToRad()(input)
"""
def __init__(self):
super(DegToRad, self).__init__()
def forward(self, input):
return deg2rad(input)
class ConvertPointsFromHomogeneous(nn.Module):
r"""Creates a transformation that converts points from homogeneous to
Euclidean space.
Args:
points (Tensor): tensor of N-dimensional points.
Returns:
Tensor: tensor of N-1-dimensional points.
Shape:
- Input: :math:`(B, D, N)` or :math:`(D, N)`
- Output: :math:`(B, D, N + 1)` or :math:`(D, N + 1)`
Examples::
>>> input = torch.rand(2, 4, 3) # BxNx3
>>> transform = tgm.ConvertPointsFromHomogeneous()
>>> output = transform(input) # BxNx2
"""
def __init__(self):
super(ConvertPointsFromHomogeneous, self).__init__()
def forward(self, input):
return convert_points_from_homogeneous(input)
class ConvertPointsToHomogeneous(nn.Module):
r"""Creates a transformation to convert points from Euclidean to
homogeneous space.
Args:
points (Tensor): tensor of N-dimensional points.
Returns:
Tensor: tensor of N+1-dimensional points.
Shape:
- Input: :math:`(B, D, N)` or :math:`(D, N)`
- Output: :math:`(B, D, N + 1)` or :math:`(D, N + 1)`
Examples::
>>> input = torch.rand(2, 4, 3) # BxNx3
>>> transform = tgm.ConvertPointsToHomogeneous()
>>> output = transform(input) # BxNx4
"""
def __init__(self):
super(ConvertPointsToHomogeneous, self).__init__()
def forward(self, input):
return convert_points_to_homogeneous(input)