-
Notifications
You must be signed in to change notification settings - Fork 1.4k
/
Copy pathtest_camera_pixels.py
264 lines (243 loc) · 11 KB
/
test_camera_pixels.py
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 (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
import unittest
import torch
from pytorch3d.renderer import (
MeshRasterizer,
NDCMultinomialRaysampler,
PerspectiveCameras,
PointsRasterizationSettings,
PointsRasterizer,
PulsarPointsRenderer,
RasterizationSettings,
)
from pytorch3d.structures import Meshes, Pointclouds
from .common_testing import TestCaseMixin
"""
PyTorch3D renderers operate in an align_corners=False manner.
This file demonstrates the pixel-perfect calculation by very simple
examples.
"""
class _CommonData:
"""
Contains data for all these tests.
- Firstly, a non-square at the origin specified in ndc space and
screen space. Principal point is in the center of the image.
Focal length is 1.0 in world space.
This camera has the identity as its world to view transformation, so
it is facing down the positive z axis with y being up and x being left.
A point on the z=1.0 focal plane has its x,y world coordinate equal to
its NDC.
- Secondly, batched together with that, is a camera with the same
focal length facing in the same direction but located so that it faces
the corner of the corner pixel of the first image, with its principal
point located at its corner, so that it maps the z=1 plane to NDC just
like the first.
- a single point self.point in world space which is located on a plane 1.0
in front from the camera which is located exactly in the center
of a known pixel (self.x, self.y), specifically with negative x and slightly
positive y, so it is in the top right quadrant of the image.
- A second batch of cameras defined in screen space which exactly match the
first ones.
So that this data can be copied for making demos, it is easiest to leave
it as a freestanding class.
"""
def __init__(self):
self.H, self.W = 249, 125
self.image_size = (self.H, self.W)
self.camera_ndc = PerspectiveCameras(
focal_length=1.0,
image_size=(self.image_size,),
in_ndc=True,
T=torch.tensor([[0.0, 0.0, 0.0], [-1.0, self.H / self.W, 0.0]]),
principal_point=((-0.0, -0.0), (1.0, -self.H / self.W)),
)
# Note how principal point is specifiied
self.camera_screen = PerspectiveCameras(
focal_length=self.W / 2.0,
principal_point=((self.W / 2.0, self.H / 2.0), (0.0, self.H)),
image_size=(self.image_size,),
T=torch.tensor([[0.0, 0.0, 0.0], [-1.0, self.H / self.W, 0.0]]),
in_ndc=False,
)
# 81 is more than half of 125, 113 is a bit less than half of 249
self.x, self.y = 81, 113
self.point = [-0.304, 0.176, 1]
# The point is in the center of pixel (81, 113)
# where pixel (0,0) is the top left.
# 81 is 38/2 pixels over the midpoint (125-1)/2=62
# and 38/125=0.304
# 113 is 22/2 pixels under the midpoint (249-1)/2=124
# and 22/125=0.176
class TestPixels(TestCaseMixin, unittest.TestCase):
def test_mesh(self):
data = _CommonData()
# Three points on the plane at unit 1 from the camera in
# world space, whose mean is the known point.
verts = torch.tensor(
[[-0.288, 0.192, 1], [-0.32, 0.192, 1], [-0.304, 0.144, 1]]
)
self.assertClose(verts.mean(0), torch.tensor(data.point))
faces = torch.LongTensor([[0, 1, 2]])
# A mesh of one triangular face whose centroid is the known point
# duplicated so it can be rendered from two cameras.
meshes = Meshes(verts=[verts], faces=[faces]).extend(2)
faces_per_pixel = 2
for camera in (data.camera_ndc, data.camera_screen):
rasterizer = MeshRasterizer(
cameras=camera,
raster_settings=RasterizationSettings(
image_size=data.image_size, faces_per_pixel=faces_per_pixel
),
)
barycentric_coords_found = rasterizer(meshes).bary_coords
self.assertTupleEqual(
barycentric_coords_found.shape,
(2,) + data.image_size + (faces_per_pixel, 3),
)
# We see that the barycentric coordinates at the expected
# pixel are (1/3, 1/3, 1/3), indicating that this pixel
# hits the centroid of the triangle.
self.assertClose(
barycentric_coords_found[:, data.y, data.x, 0],
torch.full((2, 3), 1 / 3.0),
atol=1e-5,
)
def test_pointcloud(self):
data = _CommonData()
clouds = Pointclouds(points=torch.tensor([[data.point]])).extend(2)
colorful_cloud = Pointclouds(
points=torch.tensor([[data.point]]), features=torch.ones(1, 1, 3)
).extend(2)
points_per_pixel = 2
# for camera in [data.camera_screen]:
for camera in (data.camera_ndc, data.camera_screen):
rasterizer = PointsRasterizer(
cameras=camera,
raster_settings=PointsRasterizationSettings(
image_size=data.image_size,
radius=0.0001,
points_per_pixel=points_per_pixel,
),
)
# when rasterizing we expect only one pixel to be occupied
rasterizer_output = rasterizer(clouds).idx
self.assertTupleEqual(
rasterizer_output.shape, (2,) + data.image_size + (points_per_pixel,)
)
found = torch.nonzero(rasterizer_output != -1)
self.assertTupleEqual(found.shape, (2, 4))
self.assertListEqual(found[0].tolist(), [0, data.y, data.x, 0])
self.assertListEqual(found[1].tolist(), [1, data.y, data.x, 0])
if camera.in_ndc():
# Pulsar not currently working in screen space.
pulsar_renderer = PulsarPointsRenderer(rasterizer=rasterizer)
pulsar_output = pulsar_renderer(
colorful_cloud, gamma=(0.1, 0.1), znear=(0.1, 0.1), zfar=(70, 70)
)
self.assertTupleEqual(
pulsar_output.shape, (2,) + data.image_size + (3,)
)
# Look for points rendered in the red channel only, expecting our one.
# Check the first batch element only.
# TODO: Something is odd with the second.
found = torch.nonzero(pulsar_output[0, :, :, 0])
self.assertTupleEqual(found.shape, (1, 2))
self.assertListEqual(found[0].tolist(), [data.y, data.x])
# Should be:
# found = torch.nonzero(pulsar_output[:, :, :, 0])
# self.assertTupleEqual(found.shape, (2, 3))
# self.assertListEqual(found[0].tolist(), [0, data.y, data.x])
# self.assertListEqual(found[1].tolist(), [1, data.y, data.x])
def test_raysampler(self):
data = _CommonData()
gridsampler = NDCMultinomialRaysampler(
image_width=data.W,
image_height=data.H,
n_pts_per_ray=2,
min_depth=1.0,
max_depth=2.0,
)
for camera in (data.camera_ndc, data.camera_screen):
bundle = gridsampler(camera)
self.assertTupleEqual(bundle.xys.shape, (2,) + data.image_size + (2,))
self.assertTupleEqual(
bundle.directions.shape, (2,) + data.image_size + (3,)
)
self.assertClose(
bundle.xys[:, data.y, data.x],
torch.tensor(data.point[:2]).expand(2, -1),
)
# We check only the first batch element.
# Second element varies because of camera location.
self.assertClose(
bundle.directions[0, data.y, data.x],
torch.tensor(data.point),
)
def test_camera(self):
data = _CommonData()
# Our point, plus the image center, and a corner of the image.
# Located at the focal-length distance away
points = torch.tensor([data.point, [0, 0, 1], [1, data.H / data.W, 1]])
for cameras in (data.camera_ndc, data.camera_screen):
ndc_points = cameras.transform_points_ndc(points)
screen_points = cameras.transform_points_screen(points)
screen_points_without_xyflip = cameras.transform_points_screen(
points, with_xyflip=False
)
camera_points = cameras.transform_points(points)
for batch_idx in range(2):
# NDC space agrees with the original
self.assertClose(ndc_points[batch_idx], points, atol=1e-5)
# First point in screen space is the center of our expected pixel
self.assertClose(
screen_points[batch_idx][0],
torch.tensor([data.x + 0.5, data.y + 0.5, 1.0]),
atol=1e-5,
)
# Screen coords without xyflip should have x, y that negate the non-
# flipped values, and unchanged z.
self.assertClose(
screen_points_without_xyflip[batch_idx][0],
torch.tensor([-(data.x + 0.5), -(data.y + 0.5), 1.0]),
atol=1e-5,
)
# Second point in screen space is the center of the screen
self.assertClose(
screen_points[batch_idx][1],
torch.tensor([data.W / 2.0, data.H / 2.0, 1.0]),
atol=1e-5,
)
# Third point in screen space is the corner of the screen
# (corner of corner pixels)
self.assertClose(
screen_points[batch_idx][2],
torch.tensor([0.0, 0.0, 1.0]),
atol=1e-5,
)
if cameras.in_ndc():
self.assertClose(camera_points[batch_idx], ndc_points[batch_idx])
else:
# transform_points does something strange for screen cameras
if batch_idx == 0:
wanted = torch.stack(
[
data.W - screen_points[batch_idx, :, 0],
data.H - screen_points[batch_idx, :, 1],
torch.ones(3),
],
dim=1,
)
else:
wanted = torch.stack(
[
-screen_points[batch_idx, :, 0],
2 * data.H - screen_points[batch_idx, :, 1],
torch.ones(3),
],
dim=1,
)
self.assertClose(camera_points[batch_idx], wanted)