-
Notifications
You must be signed in to change notification settings - Fork 361
/
splat_test.py
184 lines (155 loc) · 7.35 KB
/
splat_test.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
# Copyright 2020 The TensorFlow Authors
#
# 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.
"""Tests for google3.research.vision.viscam.diffren.mesh.splat."""
from absl.testing import parameterized
import tensorflow as tf
from tensorflow_graphics.rendering import rasterization_backend
from tensorflow_graphics.rendering import splat
from tensorflow_graphics.rendering import triangle_rasterizer
from tensorflow_graphics.rendering.tests import rasterization_test_utils
from tensorflow_graphics.util import test_case
@tf.function
def rasterize_image(vertices, triangles, color, camera_matrix, image_width,
image_height):
rasterized = triangle_rasterizer.rasterize(
vertices,
triangles, {'color': color},
camera_matrix, (image_width, image_height),
backend=rasterization_backend.RasterizationBackends.CPU)
return rasterized['color']
class SplatTest(test_case.TestCase):
@parameterized.parameters(([1],), ([2],), ([1, 2, 3],))
def test_batch_dimension_preserved(self, batch_shape):
"""Tests that the input batch dimension preserved."""
(vertices, triangles, attributes_dictionary, _, _, view_projection_matrix,
image_height, image_width
) = rasterization_test_utils.create_rasterizer_inputs(batch_shape)
rgba = splat.rasterize_then_splat(vertices, triangles,
attributes_dictionary,
view_projection_matrix,
(image_height, image_width),
lambda x: x['attribute1'])
tensor_batch_shape = rgba.shape.as_list()[:len(batch_shape)]
self.assertEqual(
list(batch_shape),
tensor_batch_shape,
msg='Output has batch shape {0}, but expected is {1}'.format(
tensor_batch_shape, batch_shape))
def test_rasterizes_correct_shape(self):
"""Tests that rasterize returns the expected result."""
batch_shape = []
(vertices, triangles, attributes_dictionary, _, _, view_projection_matrix,
image_height, image_width
) = rasterization_test_utils.create_rasterizer_inputs(batch_shape)
rgba = splat.rasterize_then_splat(vertices, triangles,
attributes_dictionary,
view_projection_matrix,
(image_height, image_width),
lambda x: x['attribute1'])
self.assertAllEqual(rgba.shape, (image_height, image_width, 4))
def test_two_triangle_layers(self):
"""Checks that two overlapping triangles are accumulated correctly."""
image_width = 32
image_height = 32
vertices = tf.constant(
[[[-0.2, -0.2, 0], [0.5, -0.2, 0], [0.5, 0.5, 0], [0.2, -0.2, 0.5],
[-0.5, -0.2, 0.5], [-0.5, 0.5, 0.5]]],
dtype=tf.float32)
triangles = [[0, 1, 2], [3, 5, 4]]
colors = tf.constant(
[[[0, 1.0, 0, 1.0], [0, 1.0, 0, 1.0], [0, 1.0, 0, 1.0],
[1.0, 0, 0, 1.0], [1.0, 0, 0, 1.0], [1.0, 0, 0, 1.0]]],
dtype=tf.float32)
composite, _, normalized_layers = splat.rasterize_then_splat(
vertices,
triangles, {'color': colors},
rasterization_test_utils.get_identity_view_projection_matrix(),
(image_height, image_width),
lambda x: x['color'],
num_layers=2,
return_extra_buffers=True)
baseline_image = rasterization_test_utils.load_baseline_image(
'Two_Triangles_Splat_Composite.png')
baseline_image = tf.image.resize(baseline_image,
(image_height, image_width))
images_near, error_message = rasterization_test_utils.compare_images(
self, baseline_image, composite)
self.assertTrue(images_near, msg=error_message)
for i in range(3):
baseline_image = rasterization_test_utils.load_baseline_image(
'Two_Triangles_Splat_Layer_{}.png'.format(i))
image = normalized_layers[:, i, ...]
image = tf.image.resize(
image, (512, 512), method=tf.image.ResizeMethod.NEAREST_NEIGHBOR)
images_near, error_message = rasterization_test_utils.compare_images(
self, baseline_image, image)
self.assertTrue(images_near, msg=error_message)
def test_optimize_single_triangle(self):
"""Checks that the position of a triangle can be optimized correctly.
The optimization target is a translated version of the same triangle.
Naive rasterization produces zero gradient in this case, but
rasterize-then-splat produces a useful gradient.
"""
image_width = 32
image_height = 32
initial_vertices = tf.constant([[[0, 0, 0], [0.5, 0, 0], [0.5, 0.5, 0]]],
dtype=tf.float32)
target_vertices = tf.constant(
[[[-0.25, 0, 0], [0.25, 0, 0], [0.25, 0.5, 0]]], dtype=tf.float32)
triangles = [[0, 1, 2]]
colors = tf.constant(
[[[0, 0.8, 0, 1.0], [0, 0.8, 0, 1.0], [0, 0.8, 0, 1.0]]],
dtype=tf.float32)
camera_matrix = rasterization_test_utils.get_identity_view_projection_matrix(
)
@tf.function
def render_splat(verts):
return splat.rasterize_then_splat(
verts,
triangles, {'color': colors},
camera_matrix, (image_height, image_width),
lambda x: x['color'],
num_layers=1)
# Perform a few iterations of gradient descent.
num_iters = 15
var_verts = tf.Variable(initial_vertices)
splat_loss_initial = 0.0
for i in range(num_iters):
with tf.GradientTape(persistent=True) as g:
target_image = rasterize_image(target_vertices, triangles, colors,
camera_matrix, image_width,
image_height)[0, ...]
rasterized_only_image = rasterize_image(var_verts, triangles, colors,
camera_matrix, image_width,
image_height)[0, ...]
splat_image = render_splat(var_verts)
rasterized_loss = tf.reduce_mean(
(rasterized_only_image - target_image)**2)
splat_loss = tf.reduce_mean((splat_image - target_image)**2)
rasterized_grad = g.gradient(rasterized_loss, var_verts)
splat_grad = g.gradient(splat_loss, var_verts)
if i == 0:
# Check that the rasterized-only gradient is zero, while the
# rasterize-then-splat gradient is non-zero.
self.assertAlmostEqual(tf.norm(rasterized_grad).numpy(), 0.0)
self.assertGreater(tf.norm(splat_grad).numpy(), 0.01)
splat_loss_initial = splat_loss
# Apply the gradient.
var_verts.assign_sub(splat_grad)
# Check that gradient descent reduces the loss by at least 50%.
opt_image = render_splat(var_verts)
opt_loss = tf.reduce_mean((opt_image - target_image)**2)
self.assertLess(opt_loss.numpy(), splat_loss_initial.numpy() * 0.5)
if __name__ == '__main__':
test_case.main()