test_cameras.py 37.8 KB
Newer Older
facebook-github-bot's avatar
facebook-github-bot committed
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
# Copyright (c) Facebook, Inc. and its affiliates. All rights reserved.


# Some of the code below is adapted from Soft Rasterizer (SoftRas)
#
# Copyright (c) 2017 Hiroharu Kato
# Copyright (c) 2018 Nikos Kolotouros
# Copyright (c) 2019 Shichen Liu
#
# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:
#
# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.
#
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.

import math
import unittest

31
32
33
import numpy as np
import torch
from common_testing import TestCaseMixin
Georgia Gkioxari's avatar
Georgia Gkioxari committed
34
35
36
37
from pytorch3d.renderer.cameras import OpenGLOrthographicCameras  # deprecated
from pytorch3d.renderer.cameras import OpenGLPerspectiveCameras  # deprecated
from pytorch3d.renderer.cameras import SfMOrthographicCameras  # deprecated
from pytorch3d.renderer.cameras import SfMPerspectiveCameras  # deprecated
facebook-github-bot's avatar
facebook-github-bot committed
38
from pytorch3d.renderer.cameras import (
39
    CamerasBase,
Georgia Gkioxari's avatar
Georgia Gkioxari committed
40
41
42
43
    FoVOrthographicCameras,
    FoVPerspectiveCameras,
    OrthographicCameras,
    PerspectiveCameras,
facebook-github-bot's avatar
facebook-github-bot committed
44
45
46
    camera_position_from_spherical_angles,
    get_world_to_view_transform,
    look_at_rotation,
47
    look_at_view_transform,
facebook-github-bot's avatar
facebook-github-bot committed
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
)
from pytorch3d.transforms import Transform3d
from pytorch3d.transforms.so3 import so3_exponential_map


# Naive function adapted from SoftRasterizer for test purposes.
def perspective_project_naive(points, fov=60.0):
    """
    Compute perspective projection from a given viewing angle.
    Args:
        points: (N, V, 3) representing the padded points.
        viewing angle: degrees
    Returns:
        (N, V, 3) tensor of projected points preserving the view space z
        coordinate (no z renormalization)
    """
    device = points.device
65
    halfFov = torch.tensor((fov / 2) / 180 * np.pi, dtype=torch.float32, device=device)
facebook-github-bot's avatar
facebook-github-bot committed
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
    scale = torch.tan(halfFov[None])
    scale = scale[:, None]
    z = points[:, :, 2]
    x = points[:, :, 0] / z / scale
    y = points[:, :, 1] / z / scale
    points = torch.stack((x, y, z), dim=2)
    return points


def sfm_perspective_project_naive(points, fx=1.0, fy=1.0, p0x=0.0, p0y=0.0):
    """
    Compute perspective projection using focal length and principal point.

    Args:
        points: (N, V, 3) representing the padded points.
        fx: world units
        fy: world units
        p0x: pixels
        p0y: pixels
    Returns:
        (N, V, 3) tensor of projected points.
    """
    z = points[:, :, 2]
89
90
    x = (points[:, :, 0] * fx) / z + p0x
    y = (points[:, :, 1] * fy) / z + p0y
facebook-github-bot's avatar
facebook-github-bot committed
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
    points = torch.stack((x, y, 1.0 / z), dim=2)
    return points


# Naive function adapted from SoftRasterizer for test purposes.
def orthographic_project_naive(points, scale_xyz=(1.0, 1.0, 1.0)):
    """
    Compute orthographic projection from a given angle
    Args:
        points: (N, V, 3) representing the padded points.
        scaled: (N, 3) scaling factors for each of xyz directions
    Returns:
        (N, V, 3) tensor of projected points preserving the view space z
        coordinate (no z renormalization).
    """
    if not torch.is_tensor(scale_xyz):
        scale_xyz = torch.tensor(scale_xyz)
    scale_xyz = scale_xyz.view(-1, 3)
    z = points[:, :, 2]
    x = points[:, :, 0] * scale_xyz[:, 0]
    y = points[:, :, 1] * scale_xyz[:, 1]
    points = torch.stack((x, y, z), dim=2)
    return points


Georgia Gkioxari's avatar
Georgia Gkioxari committed
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
def ndc_to_screen_points_naive(points, imsize):
    """
    Transforms points from PyTorch3D's NDC space to screen space
    Args:
        points: (N, V, 3) representing padded points
        imsize: (N, 2) image size = (width, height)
    Returns:
        (N, V, 3) tensor of transformed points
    """
    imwidth, imheight = imsize.unbind(1)
    imwidth = imwidth.view(-1, 1)
    imheight = imheight.view(-1, 1)

    x, y, z = points.unbind(2)
    x = (1.0 - x) * (imwidth - 1) / 2.0
    y = (1.0 - y) * (imheight - 1) / 2.0
    return torch.stack((x, y, z), dim=2)


Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
135
class TestCameraHelpers(TestCaseMixin, unittest.TestCase):
facebook-github-bot's avatar
facebook-github-bot committed
136
137
138
139
140
    def setUp(self) -> None:
        super().setUp()
        torch.manual_seed(42)
        np.random.seed(42)

141
142
143
144
    def test_look_at_view_transform_from_eye_point_tuple(self):
        dist = math.sqrt(2)
        elev = math.pi / 4
        azim = 0.0
Georgia Gkioxari's avatar
Georgia Gkioxari committed
145
        eye = ((0.0, 1.0, 1.0),)
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
        # using passed values for dist, elev, azim
        R, t = look_at_view_transform(dist, elev, azim, degrees=False)
        # using other values for dist, elev, azim - eye overrides
        R_eye, t_eye = look_at_view_transform(dist=3, elev=2, azim=1, eye=eye)
        # using only eye value

        R_eye_only, t_eye_only = look_at_view_transform(eye=eye)
        self.assertTrue(torch.allclose(R, R_eye, atol=2e-7))
        self.assertTrue(torch.allclose(t, t_eye, atol=2e-7))
        self.assertTrue(torch.allclose(R, R_eye_only, atol=2e-7))
        self.assertTrue(torch.allclose(t, t_eye_only, atol=2e-7))

    def test_look_at_view_transform_default_values(self):
        dist = 1.0
        elev = 0.0
        azim = 0.0
        # Using passed values for dist, elev, azim
        R, t = look_at_view_transform(dist, elev, azim)
        # Using default dist=1.0, elev=0.0, azim=0.0
        R_default, t_default = look_at_view_transform()
        # test default = passed = expected
        self.assertTrue(torch.allclose(R, R_default, atol=2e-7))
        self.assertTrue(torch.allclose(t, t_default, atol=2e-7))

170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
    def test_look_at_view_transform_non_default_at_position(self):
        dist = 1.0
        elev = 0.0
        azim = 0.0
        at = ((1, 1, 1),)
        # Using passed values for dist, elev, azim, at
        R, t = look_at_view_transform(dist, elev, azim, at=at)
        # Using default dist=1.0, elev=0.0, azim=0.0
        R_default, t_default = look_at_view_transform()
        # test default = passed = expected
        # R must be the same, t must be translated by (1,-1,1) with respect to t_default
        t_trans = torch.tensor([1, -1, 1], dtype=torch.float32).view(1, 3)
        self.assertTrue(torch.allclose(R, R_default, atol=2e-7))
        self.assertTrue(torch.allclose(t, t_default + t_trans, atol=2e-7))

facebook-github-bot's avatar
facebook-github-bot committed
185
186
187
188
    def test_camera_position_from_angles_python_scalar(self):
        dist = 2.7
        elev = 90.0
        azim = 0.0
189
190
191
        expected_position = torch.tensor([0.0, 2.7, 0.0], dtype=torch.float32).view(
            1, 3
        )
facebook-github-bot's avatar
facebook-github-bot committed
192
        position = camera_position_from_spherical_angles(dist, elev, azim)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
193
        self.assertClose(position, expected_position, atol=2e-7)
facebook-github-bot's avatar
facebook-github-bot committed
194
195
196
197
198
199
200
201
202
203

    def test_camera_position_from_angles_python_scalar_radians(self):
        dist = 2.7
        elev = math.pi / 2
        azim = 0.0
        expected_position = torch.tensor([0.0, 2.7, 0.0], dtype=torch.float32)
        expected_position = expected_position.view(1, 3)
        position = camera_position_from_spherical_angles(
            dist, elev, azim, degrees=False
        )
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
204
        self.assertClose(position, expected_position, atol=2e-7)
facebook-github-bot's avatar
facebook-github-bot committed
205
206
207
208
209

    def test_camera_position_from_angles_torch_scalars(self):
        dist = torch.tensor(2.7)
        elev = torch.tensor(0.0)
        azim = torch.tensor(90.0)
210
211
212
        expected_position = torch.tensor([2.7, 0.0, 0.0], dtype=torch.float32).view(
            1, 3
        )
facebook-github-bot's avatar
facebook-github-bot committed
213
        position = camera_position_from_spherical_angles(dist, elev, azim)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
214
        self.assertClose(position, expected_position, atol=2e-7)
facebook-github-bot's avatar
facebook-github-bot committed
215
216
217
218
219

    def test_camera_position_from_angles_mixed_scalars(self):
        dist = 2.7
        elev = torch.tensor(0.0)
        azim = 90.0
220
221
222
        expected_position = torch.tensor([2.7, 0.0, 0.0], dtype=torch.float32).view(
            1, 3
        )
facebook-github-bot's avatar
facebook-github-bot committed
223
        position = camera_position_from_spherical_angles(dist, elev, azim)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
224
        self.assertClose(position, expected_position, atol=2e-7)
facebook-github-bot's avatar
facebook-github-bot committed
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240

    def test_camera_position_from_angles_torch_scalar_grads(self):
        dist = torch.tensor(2.7, requires_grad=True)
        elev = torch.tensor(45.0, requires_grad=True)
        azim = torch.tensor(45.0)
        position = camera_position_from_spherical_angles(dist, elev, azim)
        position.sum().backward()
        self.assertTrue(hasattr(elev, "grad"))
        self.assertTrue(hasattr(dist, "grad"))
        elev_grad = elev.grad.clone()
        dist_grad = dist.grad.clone()
        elev = math.pi / 180.0 * elev.detach()
        azim = math.pi / 180.0 * azim
        grad_dist = (
            torch.cos(elev) * torch.sin(azim)
            + torch.sin(elev)
241
            + torch.cos(elev) * torch.cos(azim)
facebook-github-bot's avatar
facebook-github-bot committed
242
243
        )
        grad_elev = (
Nikhila Ravi's avatar
Nikhila Ravi committed
244
            -(torch.sin(elev)) * torch.sin(azim)
facebook-github-bot's avatar
facebook-github-bot committed
245
            + torch.cos(elev)
246
            - torch.sin(elev) * torch.cos(azim)
facebook-github-bot's avatar
facebook-github-bot committed
247
248
        )
        grad_elev = dist * (math.pi / 180.0) * grad_elev
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
249
250
        self.assertClose(elev_grad, grad_elev)
        self.assertClose(dist_grad, grad_dist)
facebook-github-bot's avatar
facebook-github-bot committed
251
252
253
254
255
256
257
258
259

    def test_camera_position_from_angles_vectors(self):
        dist = torch.tensor([2.0, 2.0])
        elev = torch.tensor([0.0, 90.0])
        azim = torch.tensor([90.0, 0.0])
        expected_position = torch.tensor(
            [[2.0, 0.0, 0.0], [0.0, 2.0, 0.0]], dtype=torch.float32
        )
        position = camera_position_from_spherical_angles(dist, elev, azim)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
260
        self.assertClose(position, expected_position, atol=2e-7)
facebook-github-bot's avatar
facebook-github-bot committed
261
262
263
264
265
266

    def test_camera_position_from_angles_vectors_broadcast(self):
        dist = torch.tensor([2.0, 3.0, 5.0])
        elev = torch.tensor([0.0])
        azim = torch.tensor([90.0])
        expected_position = torch.tensor(
267
            [[2.0, 0.0, 0.0], [3.0, 0.0, 0.0], [5.0, 0.0, 0.0]], dtype=torch.float32
facebook-github-bot's avatar
facebook-github-bot committed
268
269
        )
        position = camera_position_from_spherical_angles(dist, elev, azim)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
270
        self.assertClose(position, expected_position, atol=3e-7)
facebook-github-bot's avatar
facebook-github-bot committed
271
272
273
274
275
276

    def test_camera_position_from_angles_vectors_mixed_broadcast(self):
        dist = torch.tensor([2.0, 3.0, 5.0])
        elev = 0.0
        azim = torch.tensor(90.0)
        expected_position = torch.tensor(
277
            [[2.0, 0.0, 0.0], [3.0, 0.0, 0.0], [5.0, 0.0, 0.0]], dtype=torch.float32
facebook-github-bot's avatar
facebook-github-bot committed
278
279
        )
        position = camera_position_from_spherical_angles(dist, elev, azim)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
280
        self.assertClose(position, expected_position, atol=3e-7)
facebook-github-bot's avatar
facebook-github-bot committed
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297

    def test_camera_position_from_angles_vectors_mixed_broadcast_grads(self):
        dist = torch.tensor([2.0, 3.0, 5.0], requires_grad=True)
        elev = torch.tensor(45.0, requires_grad=True)
        azim = 45.0
        position = camera_position_from_spherical_angles(dist, elev, azim)
        position.sum().backward()
        self.assertTrue(hasattr(elev, "grad"))
        self.assertTrue(hasattr(dist, "grad"))
        elev_grad = elev.grad.clone()
        dist_grad = dist.grad.clone()
        azim = torch.tensor(azim)
        elev = math.pi / 180.0 * elev.detach()
        azim = math.pi / 180.0 * azim
        grad_dist = (
            torch.cos(elev) * torch.sin(azim)
            + torch.sin(elev)
298
            + torch.cos(elev) * torch.cos(azim)
facebook-github-bot's avatar
facebook-github-bot committed
299
300
        )
        grad_elev = (
Nikhila Ravi's avatar
Nikhila Ravi committed
301
            -(torch.sin(elev)) * torch.sin(azim)
facebook-github-bot's avatar
facebook-github-bot committed
302
            + torch.cos(elev)
303
            - torch.sin(elev) * torch.cos(azim)
facebook-github-bot's avatar
facebook-github-bot committed
304
305
        )
        grad_elev = (dist * (math.pi / 180.0) * grad_elev).sum()
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
306
307
        self.assertClose(elev_grad, grad_elev)
        self.assertClose(dist_grad, torch.full([3], grad_dist))
facebook-github-bot's avatar
facebook-github-bot committed
308
309
310
311
312
313
314
315
316
317
318
319

    def test_camera_position_from_angles_vectors_bad_broadcast(self):
        # Batch dim for broadcast must be N or 1
        dist = torch.tensor([2.0, 3.0, 5.0])
        elev = torch.tensor([0.0, 90.0])
        azim = torch.tensor([90.0])
        with self.assertRaises(ValueError):
            camera_position_from_spherical_angles(dist, elev, azim)

    def test_look_at_rotation_python_list(self):
        camera_position = [[0.0, 0.0, -1.0]]  # camera pointing along negative z
        rot_mat = look_at_rotation(camera_position)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
320
        self.assertClose(rot_mat, torch.eye(3)[None], atol=2e-7)
facebook-github-bot's avatar
facebook-github-bot committed
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346

    def test_look_at_rotation_input_fail(self):
        camera_position = [-1.0]  # expected to have xyz positions
        with self.assertRaises(ValueError):
            look_at_rotation(camera_position)

    def test_look_at_rotation_list_broadcast(self):
        # fmt: off
        camera_positions = [[0.0, 0.0, -1.0], [0.0, 0.0, 1.0]]
        rot_mats_expected = torch.tensor(
            [
                [
                    [1.0, 0.0, 0.0],
                    [0.0, 1.0, 0.0],
                    [0.0, 0.0, 1.0]
                ],
                [
                    [-1.0, 0.0,  0.0],  # noqa: E241, E201
                    [ 0.0, 1.0,  0.0],  # noqa: E241, E201
                    [ 0.0, 0.0, -1.0]   # noqa: E241, E201
                ],
            ],
            dtype=torch.float32
        )
        # fmt: on
        rot_mats = look_at_rotation(camera_positions)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
347
        self.assertClose(rot_mats, rot_mats_expected, atol=2e-7)
facebook-github-bot's avatar
facebook-github-bot committed
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371

    def test_look_at_rotation_tensor_broadcast(self):
        # fmt: off
        camera_positions = torch.tensor([
            [0.0, 0.0, -1.0],
            [0.0, 0.0,  1.0]   # noqa: E241, E201
        ], dtype=torch.float32)
        rot_mats_expected = torch.tensor(
            [
                [
                    [1.0, 0.0, 0.0],
                    [0.0, 1.0, 0.0],
                    [0.0, 0.0, 1.0]
                ],
                [
                    [-1.0, 0.0,  0.0],  # noqa: E241, E201
                    [ 0.0, 1.0,  0.0],  # noqa: E241, E201
                    [ 0.0, 0.0, -1.0]   # noqa: E241, E201
                ],
            ],
            dtype=torch.float32
        )
        # fmt: on
        rot_mats = look_at_rotation(camera_positions)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
372
        self.assertClose(rot_mats, rot_mats_expected, atol=2e-7)
facebook-github-bot's avatar
facebook-github-bot committed
373
374
375
376
377
378

    def test_look_at_rotation_tensor_grad(self):
        camera_position = torch.tensor([[0.0, 0.0, -1.0]], requires_grad=True)
        rot_mat = look_at_rotation(camera_position)
        rot_mat.sum().backward()
        self.assertTrue(hasattr(camera_position, "grad"))
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
379
380
        self.assertClose(
            camera_position.grad, torch.zeros_like(camera_position), atol=2e-7
facebook-github-bot's avatar
facebook-github-bot committed
381
382
383
384
385
386
387
388
        )

    def test_view_transform(self):
        T = torch.tensor([0.0, 0.0, -1.0], requires_grad=True).view(1, -1)
        R = look_at_rotation(T)
        RT = get_world_to_view_transform(R=R, T=T)
        self.assertTrue(isinstance(RT, Transform3d))

389
390

class TestCamerasCommon(TestCaseMixin, unittest.TestCase):
facebook-github-bot's avatar
facebook-github-bot committed
391
392
393
394
395
396
397
398
399
    def test_view_transform_class_method(self):
        T = torch.tensor([0.0, 0.0, -1.0], requires_grad=True).view(1, -1)
        R = look_at_rotation(T)
        RT = get_world_to_view_transform(R=R, T=T)
        for cam_type in (
            OpenGLPerspectiveCameras,
            OpenGLOrthographicCameras,
            SfMOrthographicCameras,
            SfMPerspectiveCameras,
Georgia Gkioxari's avatar
Georgia Gkioxari committed
400
401
402
403
            FoVOrthographicCameras,
            FoVPerspectiveCameras,
            OrthographicCameras,
            PerspectiveCameras,
facebook-github-bot's avatar
facebook-github-bot committed
404
405
406
        ):
            cam = cam_type(R=R, T=T)
            RT_class = cam.get_world_to_view_transform()
407
            self.assertTrue(torch.allclose(RT.get_matrix(), RT_class.get_matrix()))
facebook-github-bot's avatar
facebook-github-bot committed
408
409
410
411
412
413
414
415
416
417
418

        self.assertTrue(isinstance(RT, Transform3d))

    def test_get_camera_center(self, batch_size=10):
        T = torch.randn(batch_size, 3)
        R = so3_exponential_map(torch.randn(batch_size, 3) * 3.0)
        for cam_type in (
            OpenGLPerspectiveCameras,
            OpenGLOrthographicCameras,
            SfMOrthographicCameras,
            SfMPerspectiveCameras,
Georgia Gkioxari's avatar
Georgia Gkioxari committed
419
420
421
422
            FoVOrthographicCameras,
            FoVPerspectiveCameras,
            OrthographicCameras,
            PerspectiveCameras,
facebook-github-bot's avatar
facebook-github-bot committed
423
424
425
426
427
428
        ):
            cam = cam_type(R=R, T=T)
            C = cam.get_camera_center()
            C_ = -torch.bmm(R, T[:, :, None])[:, :, 0]
            self.assertTrue(torch.allclose(C, C_, atol=1e-05))

429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
    @staticmethod
    def init_random_cameras(cam_type: CamerasBase, batch_size: int):
        cam_params = {}
        T = torch.randn(batch_size, 3) * 0.03
        T[:, 2] = 4
        R = so3_exponential_map(torch.randn(batch_size, 3) * 3.0)
        cam_params = {"R": R, "T": T}
        if cam_type in (OpenGLPerspectiveCameras, OpenGLOrthographicCameras):
            cam_params["znear"] = torch.rand(batch_size) * 10 + 0.1
            cam_params["zfar"] = torch.rand(batch_size) * 4 + 1 + cam_params["znear"]
            if cam_type == OpenGLPerspectiveCameras:
                cam_params["fov"] = torch.rand(batch_size) * 60 + 30
                cam_params["aspect_ratio"] = torch.rand(batch_size) * 0.5 + 0.5
            else:
                cam_params["top"] = torch.rand(batch_size) * 0.2 + 0.9
Nikhila Ravi's avatar
Nikhila Ravi committed
444
445
                cam_params["bottom"] = -(torch.rand(batch_size)) * 0.2 - 0.9
                cam_params["left"] = -(torch.rand(batch_size)) * 0.2 - 0.9
446
                cam_params["right"] = torch.rand(batch_size) * 0.2 + 0.9
Georgia Gkioxari's avatar
Georgia Gkioxari committed
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
        elif cam_type in (FoVPerspectiveCameras, FoVOrthographicCameras):
            cam_params["znear"] = torch.rand(batch_size) * 10 + 0.1
            cam_params["zfar"] = torch.rand(batch_size) * 4 + 1 + cam_params["znear"]
            if cam_type == FoVPerspectiveCameras:
                cam_params["fov"] = torch.rand(batch_size) * 60 + 30
                cam_params["aspect_ratio"] = torch.rand(batch_size) * 0.5 + 0.5
            else:
                cam_params["max_y"] = torch.rand(batch_size) * 0.2 + 0.9
                cam_params["min_y"] = -(torch.rand(batch_size)) * 0.2 - 0.9
                cam_params["min_x"] = -(torch.rand(batch_size)) * 0.2 - 0.9
                cam_params["max_x"] = torch.rand(batch_size) * 0.2 + 0.9
        elif cam_type in (
            SfMOrthographicCameras,
            SfMPerspectiveCameras,
            OrthographicCameras,
            PerspectiveCameras,
        ):
464
465
            cam_params["focal_length"] = torch.rand(batch_size) * 10 + 0.1
            cam_params["principal_point"] = torch.randn((batch_size, 2))
Georgia Gkioxari's avatar
Georgia Gkioxari committed
466

467
468
469
470
        else:
            raise ValueError(str(cam_type))
        return cam_type(**cam_params)

Georgia Gkioxari's avatar
Georgia Gkioxari committed
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
    @staticmethod
    def init_equiv_cameras_ndc_screen(cam_type: CamerasBase, batch_size: int):
        T = torch.randn(batch_size, 3) * 0.03
        T[:, 2] = 4
        R = so3_exponential_map(torch.randn(batch_size, 3) * 3.0)
        screen_cam_params = {"R": R, "T": T}
        ndc_cam_params = {"R": R, "T": T}
        if cam_type in (OrthographicCameras, PerspectiveCameras):
            ndc_cam_params["focal_length"] = torch.rand((batch_size, 2)) * 3.0
            ndc_cam_params["principal_point"] = torch.randn((batch_size, 2))

            image_size = torch.randint(low=2, high=64, size=(batch_size, 2))
            screen_cam_params["image_size"] = image_size
            screen_cam_params["focal_length"] = (
                ndc_cam_params["focal_length"] * image_size / 2.0
            )
            screen_cam_params["principal_point"] = (
                (1.0 - ndc_cam_params["principal_point"]) * image_size / 2.0
            )
        else:
            raise ValueError(str(cam_type))
        return cam_type(**ndc_cam_params), cam_type(**screen_cam_params)

494
495
496
497
498
499
500
501
502
503
504
    def test_unproject_points(self, batch_size=50, num_points=100):
        """
        Checks that an unprojection of a randomly projected point cloud
        stays the same.
        """

        for cam_type in (
            SfMOrthographicCameras,
            OpenGLPerspectiveCameras,
            OpenGLOrthographicCameras,
            SfMPerspectiveCameras,
Georgia Gkioxari's avatar
Georgia Gkioxari committed
505
506
507
508
            FoVOrthographicCameras,
            FoVPerspectiveCameras,
            OrthographicCameras,
            PerspectiveCameras,
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
        ):
            # init the cameras
            cameras = TestCamerasCommon.init_random_cameras(cam_type, batch_size)
            # xyz - the ground truth point cloud
            xyz = torch.randn(batch_size, num_points, 3) * 0.3
            # xyz in camera coordinates
            xyz_cam = cameras.get_world_to_view_transform().transform_points(xyz)
            # depth = z-component of xyz_cam
            depth = xyz_cam[:, :, 2:]
            # project xyz
            xyz_proj = cameras.transform_points(xyz)
            xy, cam_depth = xyz_proj.split(2, dim=2)
            # input to the unprojection function
            xy_depth = torch.cat((xy, depth), dim=2)

            for to_world in (False, True):
                if to_world:
                    matching_xyz = xyz
                else:
                    matching_xyz = xyz_cam

Georgia Gkioxari's avatar
Georgia Gkioxari committed
530
                # if we have FoV (= OpenGL) cameras
531
                # test for scaled_depth_input=True/False
Georgia Gkioxari's avatar
Georgia Gkioxari committed
532
533
534
535
536
537
                if cam_type in (
                    OpenGLPerspectiveCameras,
                    OpenGLOrthographicCameras,
                    FoVPerspectiveCameras,
                    FoVOrthographicCameras,
                ):
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
                    for scaled_depth_input in (True, False):
                        if scaled_depth_input:
                            xy_depth_ = xyz_proj
                        else:
                            xy_depth_ = xy_depth
                        xyz_unproj = cameras.unproject_points(
                            xy_depth_,
                            world_coordinates=to_world,
                            scaled_depth_input=scaled_depth_input,
                        )
                        self.assertTrue(
                            torch.allclose(xyz_unproj, matching_xyz, atol=1e-4)
                        )
                else:
                    xyz_unproj = cameras.unproject_points(
                        xy_depth, world_coordinates=to_world
                    )
                    self.assertTrue(torch.allclose(xyz_unproj, matching_xyz, atol=1e-4))

Georgia Gkioxari's avatar
Georgia Gkioxari committed
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
    def test_project_points_screen(self, batch_size=50, num_points=100):
        """
        Checks that an unprojection of a randomly projected point cloud
        stays the same.
        """

        for cam_type in (
            OpenGLOrthographicCameras,
            OpenGLPerspectiveCameras,
            SfMOrthographicCameras,
            SfMPerspectiveCameras,
            FoVOrthographicCameras,
            FoVPerspectiveCameras,
            OrthographicCameras,
            PerspectiveCameras,
        ):

            # init the cameras
            cameras = TestCamerasCommon.init_random_cameras(cam_type, batch_size)
            # xyz - the ground truth point cloud
            xyz = torch.randn(batch_size, num_points, 3) * 0.3
            # image size
            image_size = torch.randint(low=2, high=64, size=(batch_size, 2))
            # project points
            xyz_project_ndc = cameras.transform_points(xyz)
            xyz_project_screen = cameras.transform_points_screen(xyz, image_size)
            # naive
            xyz_project_screen_naive = ndc_to_screen_points_naive(
                xyz_project_ndc, image_size
            )
            self.assertClose(xyz_project_screen, xyz_project_screen_naive)

    def test_equiv_project_points(self, batch_size=50, num_points=100):
        """
        Checks that NDC and screen cameras project points to ndc correctly.
        Applies only to OrthographicCameras and PerspectiveCameras.
        """
        for cam_type in (OrthographicCameras, PerspectiveCameras):
            # init the cameras
            (
                ndc_cameras,
                screen_cameras,
            ) = TestCamerasCommon.init_equiv_cameras_ndc_screen(cam_type, batch_size)
            # xyz - the ground truth point cloud
            xyz = torch.randn(batch_size, num_points, 3) * 0.3
            # project points
            xyz_ndc_cam = ndc_cameras.transform_points(xyz)
            xyz_screen_cam = screen_cameras.transform_points(xyz)
            self.assertClose(xyz_ndc_cam, xyz_screen_cam, atol=1e-6)

607
608
609
610
611
612
613
614
615
    def test_clone(self, batch_size: int = 10):
        """
        Checks the clone function of the cameras.
        """
        for cam_type in (
            SfMOrthographicCameras,
            OpenGLPerspectiveCameras,
            OpenGLOrthographicCameras,
            SfMPerspectiveCameras,
Georgia Gkioxari's avatar
Georgia Gkioxari committed
616
617
618
619
            FoVOrthographicCameras,
            FoVPerspectiveCameras,
            OrthographicCameras,
            PerspectiveCameras,
620
621
622
623
624
625
626
627
628
629
630
631
632
633
        ):
            cameras = TestCamerasCommon.init_random_cameras(cam_type, batch_size)
            cameras = cameras.to(torch.device("cpu"))
            cameras_clone = cameras.clone()

            for var in cameras.__dict__.keys():
                val = getattr(cameras, var)
                val_clone = getattr(cameras_clone, var)
                if torch.is_tensor(val):
                    self.assertClose(val, val_clone)
                    self.assertSeparate(val, val_clone)
                else:
                    self.assertTrue(val == val_clone)

facebook-github-bot's avatar
facebook-github-bot committed
634

Georgia Gkioxari's avatar
Georgia Gkioxari committed
635
636
637
638
639
640
############################################################
#                FoVPerspective Camera                     #
############################################################


class TestFoVPerspectiveProjection(TestCaseMixin, unittest.TestCase):
facebook-github-bot's avatar
facebook-github-bot committed
641
642
643
    def test_perspective(self):
        far = 10.0
        near = 1.0
Georgia Gkioxari's avatar
Georgia Gkioxari committed
644
        cameras = FoVPerspectiveCameras(znear=near, zfar=far, fov=60.0)
facebook-github-bot's avatar
facebook-github-bot committed
645
646
647
648
649
650
651
652
653
        P = cameras.get_projection_transform()
        # vertices are at the far clipping plane so z gets mapped to 1.
        vertices = torch.tensor([1, 2, far], dtype=torch.float32)
        projected_verts = torch.tensor(
            [np.sqrt(3) / far, 2 * np.sqrt(3) / far, 1.0], dtype=torch.float32
        )
        vertices = vertices[None, None, :]
        v1 = P.transform_points(vertices)
        v2 = perspective_project_naive(vertices, fov=60.0)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
654
655
656
        self.assertClose(v1[..., :2], v2[..., :2])
        self.assertClose(far * v1[..., 2], v2[..., 2])
        self.assertClose(v1.squeeze(), projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
657
658
659
660
661
662
663
664

        # vertices are at the near clipping plane so z gets mapped to 0.0.
        vertices[..., 2] = near
        projected_verts = torch.tensor(
            [np.sqrt(3) / near, 2 * np.sqrt(3) / near, 0.0], dtype=torch.float32
        )
        v1 = P.transform_points(vertices)
        v2 = perspective_project_naive(vertices, fov=60.0)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
665
666
        self.assertClose(v1[..., :2], v2[..., :2])
        self.assertClose(v1.squeeze(), projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
667
668

    def test_perspective_kwargs(self):
Georgia Gkioxari's avatar
Georgia Gkioxari committed
669
        cameras = FoVPerspectiveCameras(znear=5.0, zfar=100.0, fov=0.0)
facebook-github-bot's avatar
facebook-github-bot committed
670
671
672
673
674
675
676
677
678
        # Override defaults by passing in values to get_projection_transform
        far = 10.0
        P = cameras.get_projection_transform(znear=1.0, zfar=far, fov=60.0)
        vertices = torch.tensor([1, 2, far], dtype=torch.float32)
        projected_verts = torch.tensor(
            [np.sqrt(3) / far, 2 * np.sqrt(3) / far, 1.0], dtype=torch.float32
        )
        vertices = vertices[None, None, :]
        v1 = P.transform_points(vertices)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
679
        self.assertClose(v1.squeeze(), projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
680
681
682
683
684

    def test_perspective_mixed_inputs_broadcast(self):
        far = torch.tensor([10.0, 20.0], dtype=torch.float32)
        near = 1.0
        fov = torch.tensor(60.0)
Georgia Gkioxari's avatar
Georgia Gkioxari committed
685
        cameras = FoVPerspectiveCameras(znear=near, zfar=far, fov=fov)
facebook-github-bot's avatar
facebook-github-bot committed
686
687
688
        P = cameras.get_projection_transform()
        vertices = torch.tensor([1, 2, 10], dtype=torch.float32)
        z1 = 1.0  # vertices at far clipping plane so z = 1.0
Nikhila Ravi's avatar
Nikhila Ravi committed
689
        z2 = (20.0 / (20.0 - 1.0) * 10.0 + -20.0 / (20.0 - 1.0)) / 10.0
facebook-github-bot's avatar
facebook-github-bot committed
690
691
692
693
694
695
696
697
698
699
        projected_verts = torch.tensor(
            [
                [np.sqrt(3) / 10.0, 2 * np.sqrt(3) / 10.0, z1],
                [np.sqrt(3) / 10.0, 2 * np.sqrt(3) / 10.0, z2],
            ],
            dtype=torch.float32,
        )
        vertices = vertices[None, None, :]
        v1 = P.transform_points(vertices)
        v2 = perspective_project_naive(vertices, fov=60.0)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
700
701
        self.assertClose(v1[..., :2], torch.cat([v2, v2])[..., :2])
        self.assertClose(v1.squeeze(), projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
702
703
704
705
706

    def test_perspective_mixed_inputs_grad(self):
        far = torch.tensor([10.0])
        near = 1.0
        fov = torch.tensor(60.0, requires_grad=True)
Georgia Gkioxari's avatar
Georgia Gkioxari committed
707
        cameras = FoVPerspectiveCameras(znear=near, zfar=far, fov=fov)
facebook-github-bot's avatar
facebook-github-bot committed
708
709
710
711
712
713
714
715
716
717
718
        P = cameras.get_projection_transform()
        vertices = torch.tensor([1, 2, 10], dtype=torch.float32)
        vertices_batch = vertices[None, None, :]
        v1 = P.transform_points(vertices_batch).squeeze()
        v1.sum().backward()
        self.assertTrue(hasattr(fov, "grad"))
        fov_grad = fov.grad.clone()
        half_fov_rad = (math.pi / 180.0) * fov.detach() / 2.0
        grad_cotan = -(1.0 / (torch.sin(half_fov_rad) ** 2.0) * 1 / 2.0)
        grad_fov = (math.pi / 180.0) * grad_cotan
        grad_fov = (vertices[0] + vertices[1]) * grad_fov / 10.0
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
719
        self.assertClose(fov_grad, grad_fov)
facebook-github-bot's avatar
facebook-github-bot committed
720
721
722

    def test_camera_class_init(self):
        device = torch.device("cuda:0")
Georgia Gkioxari's avatar
Georgia Gkioxari committed
723
        cam = FoVPerspectiveCameras(znear=10.0, zfar=(100.0, 200.0))
facebook-github-bot's avatar
facebook-github-bot committed
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741

        # Check broadcasting
        self.assertTrue(cam.znear.shape == (2,))
        self.assertTrue(cam.zfar.shape == (2,))

        # update znear element 1
        cam[1].znear = 20.0
        self.assertTrue(cam.znear[1] == 20.0)

        # Get item and get value
        c0 = cam[0]
        self.assertTrue(c0.zfar == 100.0)

        # Test to
        new_cam = cam.to(device=device)
        self.assertTrue(new_cam.device == device)

    def test_get_full_transform(self):
Georgia Gkioxari's avatar
Georgia Gkioxari committed
742
        cam = FoVPerspectiveCameras()
facebook-github-bot's avatar
facebook-github-bot committed
743
744
745
746
        T = torch.tensor([0.0, 0.0, 1.0]).view(1, -1)
        R = look_at_rotation(T)
        P = cam.get_full_projection_transform(R=R, T=T)
        self.assertTrue(isinstance(P, Transform3d))
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
747
748
        self.assertClose(cam.R, R)
        self.assertClose(cam.T, T)
facebook-github-bot's avatar
facebook-github-bot committed
749
750
751
752
753

    def test_transform_points(self):
        # Check transform_points methods works with default settings for
        # RT and P
        far = 10.0
Georgia Gkioxari's avatar
Georgia Gkioxari committed
754
        cam = FoVPerspectiveCameras(znear=1.0, zfar=far, fov=60.0)
facebook-github-bot's avatar
facebook-github-bot committed
755
756
757
758
759
760
761
        points = torch.tensor([1, 2, far], dtype=torch.float32)
        points = points.view(1, 1, 3).expand(5, 10, -1)
        projected_points = torch.tensor(
            [np.sqrt(3) / far, 2 * np.sqrt(3) / far, 1.0], dtype=torch.float32
        )
        projected_points = projected_points.view(1, 1, 3).expand(5, 10, -1)
        new_points = cam.transform_points(points)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
762
        self.assertClose(new_points, projected_points)
facebook-github-bot's avatar
facebook-github-bot committed
763
764


Georgia Gkioxari's avatar
Georgia Gkioxari committed
765
766
767
768
769
770
############################################################
#                FoVOrthographic Camera                    #
############################################################


class TestFoVOrthographicProjection(TestCaseMixin, unittest.TestCase):
facebook-github-bot's avatar
facebook-github-bot committed
771
772
773
    def test_orthographic(self):
        far = 10.0
        near = 1.0
Georgia Gkioxari's avatar
Georgia Gkioxari committed
774
        cameras = FoVOrthographicCameras(znear=near, zfar=far)
facebook-github-bot's avatar
facebook-github-bot committed
775
776
777
778
779
780
781
        P = cameras.get_projection_transform()

        vertices = torch.tensor([1, 2, far], dtype=torch.float32)
        projected_verts = torch.tensor([1, 2, 1], dtype=torch.float32)
        vertices = vertices[None, None, :]
        v1 = P.transform_points(vertices)
        v2 = orthographic_project_naive(vertices)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
782
783
        self.assertClose(v1[..., :2], v2[..., :2])
        self.assertClose(v1.squeeze(), projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
784
785
786
787
788

        vertices[..., 2] = near
        projected_verts[2] = 0.0
        v1 = P.transform_points(vertices)
        v2 = orthographic_project_naive(vertices)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
789
790
        self.assertClose(v1[..., :2], v2[..., :2])
        self.assertClose(v1.squeeze(), projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
791
792
793
794
795
796
797
798

    def test_orthographic_scaled(self):
        vertices = torch.tensor([1, 2, 0.5], dtype=torch.float32)
        vertices = vertices[None, None, :]
        scale = torch.tensor([[2.0, 0.5, 20]])
        # applying the scale puts the z coordinate at the far clipping plane
        # so the z is mapped to 1.0
        projected_verts = torch.tensor([2, 1, 1], dtype=torch.float32)
Georgia Gkioxari's avatar
Georgia Gkioxari committed
799
        cameras = FoVOrthographicCameras(znear=1.0, zfar=10.0, scale_xyz=scale)
facebook-github-bot's avatar
facebook-github-bot committed
800
801
802
        P = cameras.get_projection_transform()
        v1 = P.transform_points(vertices)
        v2 = orthographic_project_naive(vertices, scale)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
803
804
        self.assertClose(v1[..., :2], v2[..., :2])
        self.assertClose(v1, projected_verts[None, None])
facebook-github-bot's avatar
facebook-github-bot committed
805
806

    def test_orthographic_kwargs(self):
Georgia Gkioxari's avatar
Georgia Gkioxari committed
807
        cameras = FoVOrthographicCameras(znear=5.0, zfar=100.0)
facebook-github-bot's avatar
facebook-github-bot committed
808
809
810
811
812
813
        far = 10.0
        P = cameras.get_projection_transform(znear=1.0, zfar=far)
        vertices = torch.tensor([1, 2, far], dtype=torch.float32)
        projected_verts = torch.tensor([1, 2, 1], dtype=torch.float32)
        vertices = vertices[None, None, :]
        v1 = P.transform_points(vertices)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
814
        self.assertClose(v1.squeeze(), projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
815
816
817
818

    def test_orthographic_mixed_inputs_broadcast(self):
        far = torch.tensor([10.0, 20.0])
        near = 1.0
Georgia Gkioxari's avatar
Georgia Gkioxari committed
819
        cameras = FoVOrthographicCameras(znear=near, zfar=far)
facebook-github-bot's avatar
facebook-github-bot committed
820
821
        P = cameras.get_projection_transform()
        vertices = torch.tensor([1.0, 2.0, 10.0], dtype=torch.float32)
Nikhila Ravi's avatar
Nikhila Ravi committed
822
        z2 = 1.0 / (20.0 - 1.0) * 10.0 + -1.0 / (20.0 - 1.0)
facebook-github-bot's avatar
facebook-github-bot committed
823
824
825
826
827
828
        projected_verts = torch.tensor(
            [[1.0, 2.0, 1.0], [1.0, 2.0, z2]], dtype=torch.float32
        )
        vertices = vertices[None, None, :]
        v1 = P.transform_points(vertices)
        v2 = orthographic_project_naive(vertices)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
829
830
        self.assertClose(v1[..., :2], torch.cat([v2, v2])[..., :2])
        self.assertClose(v1.squeeze(), projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
831
832
833
834
835

    def test_orthographic_mixed_inputs_grad(self):
        far = torch.tensor([10.0])
        near = 1.0
        scale = torch.tensor([[1.0, 1.0, 1.0]], requires_grad=True)
Georgia Gkioxari's avatar
Georgia Gkioxari committed
836
        cameras = FoVOrthographicCameras(znear=near, zfar=far, scale_xyz=scale)
facebook-github-bot's avatar
facebook-github-bot committed
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
        P = cameras.get_projection_transform()
        vertices = torch.tensor([1.0, 2.0, 10.0], dtype=torch.float32)
        vertices_batch = vertices[None, None, :]
        v1 = P.transform_points(vertices_batch)
        v1.sum().backward()
        self.assertTrue(hasattr(scale, "grad"))
        scale_grad = scale.grad.clone()
        grad_scale = torch.tensor(
            [
                [
                    vertices[0] * P._matrix[:, 0, 0],
                    vertices[1] * P._matrix[:, 1, 1],
                    vertices[2] * P._matrix[:, 2, 2],
                ]
            ]
        )
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
853
        self.assertClose(scale_grad, grad_scale)
facebook-github-bot's avatar
facebook-github-bot committed
854
855


Georgia Gkioxari's avatar
Georgia Gkioxari committed
856
857
858
859
860
861
############################################################
#                Orthographic Camera                       #
############################################################


class TestOrthographicProjection(TestCaseMixin, unittest.TestCase):
facebook-github-bot's avatar
facebook-github-bot committed
862
    def test_orthographic(self):
Georgia Gkioxari's avatar
Georgia Gkioxari committed
863
        cameras = OrthographicCameras()
facebook-github-bot's avatar
facebook-github-bot committed
864
865
866
867
868
869
870
        P = cameras.get_projection_transform()

        vertices = torch.randn([3, 4, 3], dtype=torch.float32)
        projected_verts = vertices.clone()
        v1 = P.transform_points(vertices)
        v2 = orthographic_project_naive(vertices)

Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
871
872
        self.assertClose(v1[..., :2], v2[..., :2])
        self.assertClose(v1, projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
873
874
875
876
877

    def test_orthographic_scaled(self):
        focal_length_x = 10.0
        focal_length_y = 15.0

Georgia Gkioxari's avatar
Georgia Gkioxari committed
878
        cameras = OrthographicCameras(focal_length=((focal_length_x, focal_length_y),))
facebook-github-bot's avatar
facebook-github-bot committed
879
880
881
882
883
884
885
886
887
888
889
        P = cameras.get_projection_transform()

        vertices = torch.randn([3, 4, 3], dtype=torch.float32)
        projected_verts = vertices.clone()
        projected_verts[:, :, 0] *= focal_length_x
        projected_verts[:, :, 1] *= focal_length_y
        v1 = P.transform_points(vertices)
        v2 = orthographic_project_naive(
            vertices, scale_xyz=(focal_length_x, focal_length_y, 1.0)
        )
        v3 = cameras.transform_points(vertices)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
890
891
892
        self.assertClose(v1[..., :2], v2[..., :2])
        self.assertClose(v3[..., :2], v2[..., :2])
        self.assertClose(v1, projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
893
894

    def test_orthographic_kwargs(self):
Georgia Gkioxari's avatar
Georgia Gkioxari committed
895
        cameras = OrthographicCameras(focal_length=5.0, principal_point=((2.5, 2.5),))
facebook-github-bot's avatar
facebook-github-bot committed
896
897
898
899
900
901
902
903
904
        P = cameras.get_projection_transform(
            focal_length=2.0, principal_point=((2.5, 3.5),)
        )
        vertices = torch.randn([3, 4, 3], dtype=torch.float32)
        projected_verts = vertices.clone()
        projected_verts[:, :, :2] *= 2.0
        projected_verts[:, :, 0] += 2.5
        projected_verts[:, :, 1] += 3.5
        v1 = P.transform_points(vertices)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
905
        self.assertClose(v1, projected_verts)
facebook-github-bot's avatar
facebook-github-bot committed
906
907


Georgia Gkioxari's avatar
Georgia Gkioxari committed
908
909
910
911
912
913
############################################################
#                Perspective Camera                        #
############################################################


class TestPerspectiveProjection(TestCaseMixin, unittest.TestCase):
facebook-github-bot's avatar
facebook-github-bot committed
914
    def test_perspective(self):
Georgia Gkioxari's avatar
Georgia Gkioxari committed
915
        cameras = PerspectiveCameras()
facebook-github-bot's avatar
facebook-github-bot committed
916
917
918
919
920
        P = cameras.get_projection_transform()

        vertices = torch.randn([3, 4, 3], dtype=torch.float32)
        v1 = P.transform_points(vertices)
        v2 = sfm_perspective_project_naive(vertices)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
921
        self.assertClose(v1, v2)
facebook-github-bot's avatar
facebook-github-bot committed
922
923
924
925
926
927
928

    def test_perspective_scaled(self):
        focal_length_x = 10.0
        focal_length_y = 15.0
        p0x = 15.0
        p0y = 30.0

Georgia Gkioxari's avatar
Georgia Gkioxari committed
929
        cameras = PerspectiveCameras(
facebook-github-bot's avatar
facebook-github-bot committed
930
931
932
933
934
935
936
937
938
939
940
            focal_length=((focal_length_x, focal_length_y),),
            principal_point=((p0x, p0y),),
        )
        P = cameras.get_projection_transform()

        vertices = torch.randn([3, 4, 3], dtype=torch.float32)
        v1 = P.transform_points(vertices)
        v2 = sfm_perspective_project_naive(
            vertices, fx=focal_length_x, fy=focal_length_y, p0x=p0x, p0y=p0y
        )
        v3 = cameras.transform_points(vertices)
Jeremy Reizenstein's avatar
Jeremy Reizenstein committed
941
942
        self.assertClose(v1, v2)
        self.assertClose(v3[..., :2], v2[..., :2])
facebook-github-bot's avatar
facebook-github-bot committed
943
944

    def test_perspective_kwargs(self):
Georgia Gkioxari's avatar
Georgia Gkioxari committed
945
        cameras = PerspectiveCameras(focal_length=5.0, principal_point=((2.5, 2.5),))
facebook-github-bot's avatar
facebook-github-bot committed
946
947
948
949
950
        P = cameras.get_projection_transform(
            focal_length=2.0, principal_point=((2.5, 3.5),)
        )
        vertices = torch.randn([3, 4, 3], dtype=torch.float32)
        v1 = P.transform_points(vertices)
951
        v2 = sfm_perspective_project_naive(vertices, fx=2.0, fy=2.0, p0x=2.5, p0y=3.5)
952
        self.assertClose(v1, v2, atol=1e-6)