File size: 8,473 Bytes
25a49bc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
# Copyright 2023 Google LLC
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# Author: spopov@google.com (Stefan Popov)
"""Functions to compute transformation matrices."""

import torch as t
from torch.nn import functional as F


def scale(v: t.Tensor) -> t.Tensor:
  """Computes homogeneous scale matrices from scale vectors.

  Args:
    v: Scale vectors, `float32[B*, N]`

  Returns:
    Scale matrices, `float32[B*, N+1, N+1]`
  """
  v = t.as_tensor(v, dtype=t.float32)
  batch_dims = v.shape[:-1]
  v = v.reshape([-1, (v.shape[-1])])

  index_batch_flat = t.arange(v.shape[0], dtype=t.int64, device=v.device)
  index_diag = t.arange(v.shape[1], dtype=t.int64, device=v.device)
  index_batch, index_diag = t.meshgrid(index_batch_flat, index_diag,
                                       indexing="ij")
  index_batch = index_batch.reshape([-1])
  index_diag = index_diag.reshape([-1])

  result = v.new_zeros([v.shape[0], v.shape[1] + 1, v.shape[1] + 1])
  result[index_batch, index_diag, index_diag] = v.reshape([-1])
  result[index_batch_flat, v.shape[-1], v.shape[-1]] = 1
  result = result.reshape(batch_dims + result.shape[-2:])
  return result


def translate(v: t.Tensor) -> t.Tensor:
  """Computes a homogeneous translation matrices from translation vectors.

  Args:
    v: Translation vectors, `float32[B*, N]`

  Returns:
    Translation matrices, `float32[B*, N+1, N+1]`
  """
  result = t.as_tensor(v, dtype=t.float32)
  dimensions = result.shape[-1]
  result = result[..., None, :].transpose(-1, -2)
  result = t.constant_pad_nd(result, [dimensions, 0, 0, 1])
  id_matrix = t.diag(result.new_ones([dimensions + 1]))
  id_matrix = id_matrix.expand_as(result)
  result = result + id_matrix
  return result


def rotate(
    angle: t.Tensor,
    axis: t.Tensor,
) -> t.Tensor:
  """Computes a 3D rotation matrices from angle and axis inputs.

  The formula used in this function is explained here:
  https://en.wikipedia.org/wiki/Rotation_matrix#Conversion_from_and_to_axis–angle

  Args:
    angle: The rotation angles, `float32[B*]`
    axis: The rotation axes, `float32[B*, 3]`

  Returns:
    The rotation matrices, `float32[B*, 4, 4]`
  """
  axis = t.as_tensor(axis, dtype=t.float32)
  angle = t.as_tensor(angle, dtype=t.float32)
  axis = F.normalize(axis, dim=-1)
  sin_axis = t.sin(angle)[..., None] * axis
  cos_angle = t.cos(angle)
  cos1_axis = (1.0 - cos_angle)[..., None] * axis
  _, axis_y, axis_z = t.unbind(axis, dim=-1)
  cos1_axis_x, cos1_axis_y, _ = t.unbind(cos1_axis, dim=-1)
  sin_axis_x, sin_axis_y, sin_axis_z = t.unbind(sin_axis, dim=-1)
  tmp = cos1_axis_x * axis_y
  m01 = tmp - sin_axis_z
  m10 = tmp + sin_axis_z
  tmp = cos1_axis_x * axis_z
  m02 = tmp + sin_axis_y
  m20 = tmp - sin_axis_y
  tmp = cos1_axis_y * axis_z
  m12 = tmp - sin_axis_x
  m21 = tmp + sin_axis_x
  zero = t.zeros_like(m01)
  one = t.ones_like(m01)
  diag = cos1_axis * axis + cos_angle[..., None]
  diag_x, diag_y, diag_z = t.unbind(diag, dim=-1)

  matrix = t.stack((diag_x, m01, m02, zero, m10, diag_y, m12, zero, m20, m21,
                    diag_z, zero, zero, zero, zero, one), dim=-1)
  output_shape = axis.shape[:-1] + (4, 4)
  result = matrix.reshape(output_shape)
  return result


def transform_points_homogeneous(points: t.Tensor, matrix: t.Tensor,
                                 w: float) -> t.Tensor:
  """Transforms 3D points with a homogeneous matrix.

  Args:
    points: The points to transform, `float32[B*, N, 3]`
    matrix: The transformation matrices, `float32[B*, 4, 4]`
    w: The W value to add to the points to make them homogeneous. Should be 1
      for affine points and 0 for vectors.

  Returns:
    The transformed points in homogeneous space (with a 4th coordinate),
    `float32[B*, N, 4]`
  """
  batch_dims = points.shape[:-2]
  # Fold all batch dimensions into a single one
  points = points.reshape([-1] + list(points.shape[-2:]))
  matrix = matrix.reshape([-1] + list(matrix.shape[-2:]))

  points = t.constant_pad_nd(points, [0, 1], value=w)
  result = t.einsum("bnm,bvm->bvn", matrix, points)
  result = result.reshape(batch_dims + result.shape[-2:])

  return result


def transform_mesh(mesh: t.Tensor, matrix: t.Tensor,
                   vertices_are_points=True) -> t.Tensor:
  """Transforms a single 3D mesh.

  Args:
    mesh: The mesh's triangle vertices, `float32[B*, N, 3, 3]`
    matrix: The transformation matrix, `float32[B*, 4, 4]`
    vertices_are_points: Whether to interpret the vertices as affine points
      or vectors.

  Returns:
    The transformed mesh, `float32[B*, N, 3, 3]`
  """
  original_shape = mesh.shape
  mesh = mesh.reshape([-1, mesh.shape[-3] * 3, 3])
  matrix = matrix.reshape([-1, 4, 4])
  w = 1 if vertices_are_points else 0
  mesh = transform_points_homogeneous(mesh, matrix, w=w)
  if vertices_are_points:
    mesh = mesh[..., :3] / mesh[..., 3:4]
  else:
    mesh = mesh[..., :3]

  return mesh.reshape(original_shape)


def transform_points(points: t.Tensor, matrix: t.Tensor) -> t.Tensor:
  """Transforms points.
  Args:
    points: The points to transform, `float32[B*, N, 3]`
    matrix: Transformation matrices, `float32[B*, 4, 4]`
  Result:
    The transformed points, `float32[B*, N, 3]`
  """
  result = transform_points_homogeneous(points, matrix, w=1)
  result = result[..., :3] / result[..., 3:4]
  return result


def chain(transforms: list[t.Tensor], reverse=True) -> t.Tensor:
  """Chains transformations expressed as matrices.

  Args:
    transforms: The list of transformations to chain
    reverse: The order in which transformations are applied. If true, the last
      transformation is applied first (which matches matrix multiplication
      order). False matches natural order, where the first transformation is
      applied first.

  Returns:
    Matrix combining all transformations.

  """
  assert transforms
  if not reverse:
    transforms = transforms[::-1]
  result = transforms[0]
  for transform in transforms[1:]:
    result = result @ transform
  return result


def gl_projection_matrix_from_intrinsics(  #
    width: t.Tensor, height: t.Tensor, fx: t.Tensor, fy: t.Tensor, cx: t.Tensor,
    cy: t.Tensor, znear: float = 0.001, zfar: float = 20.) -> t.Tensor:
  """Computes the camera projection matrix for rendering square images.

  Args:
    width: Image's `width`, `float32[B*]`.
    height: Image's `heigh`t,` float32[B*]`.
    fx: Camera's `fx`, `float32[B*]`.
    fy: Camera's `fy`, `float32[B*]`.
    cx: Camera's `cx`, `float32[B*]`.
    cy: Camera's `cy`, `float32[B*]`.
    znear: The near plane location.
    zfar: The far plane location.

  Returns:
    World to OpenGL's normalized device coordinates transformation matrices,
    `float32[B*, 4, 4]`.
  """

  z = t.zeros_like(t.as_tensor(fx))
  o = t.ones_like(z)
  zn = znear * o
  zf = zfar * o
  # yapf: disable
  result = [
      2 * fx / width, z, 2 * (cx / width) - 1, z,
      z, 2 * fy / height, 2 * (cy / height) - 1, z,
      z, z, (zf + zn) / (zf - zn), -2 * zn * zf / (zf - zn),
      z, z, o, z
  ]
  # yapf: enable
  result = t.stack([t.as_tensor(v, dtype=t.float32)
                    for v in result]).reshape((4, 4) + z.shape)

  result = result.permute(tuple(range(len(result.shape)))[2:] + (0, 1))
  return result


def quaternion_to_rotation_matrix(q: t.Tensor) -> t.Tensor:
  """Computes a rotation matrix from a quaternion.

  Args:
    q: Rotation quaternions, float32[B*, 4]

  Returns:
    Rotation matrices, float32[B, 4, 4]

  """
  q = t.as_tensor(q, dtype=t.float32)
  w, x, y, z = t.unbind(q, dim=-1)
  zz = t.zeros_like(z)
  oo = t.ones_like(z)
  s = 2.0 / (q * q).sum(dim=-1)
  # yapf: disable
  return t.stack([
      1 - s * (y ** 2 + z ** 2), s * (x * y - z * w), s * (x * z + y * w), zz,
      s * (x * y + z * w), 1 - s * (x ** 2 + z ** 2), s * (y * z - x * w), zz,
      s * (x * z - y * w), s * (y * z + x * w), 1 - s * (x ** 2 + y ** 2), zz,
      zz, zz, zz, oo
  ], dim=-1).reshape(q.shape[:-1] + (4, 4))
  # yapf: enable