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
|
# ----------------------------------------------------------------------------
# - Open3D: www.open3d.org -
# ----------------------------------------------------------------------------
# Copyright (c) 2018-2024 www.open3d.org
# SPDX-License-Identifier: MIT
# ----------------------------------------------------------------------------
import open3d as o3d
import numpy as np
import tempfile
import os
def test_apply_depth_noise_model():
class Simulator:
"""Original implementation of the simulator:
http://redwood-data.org/indoor/data/simdepth.py
"""
def loaddistmodel(self, fname):
data = np.loadtxt(fname, comments='%', skiprows=5)
dist = np.empty([80, 80, 5])
for y in range(0, 80):
for x in range(0, 80):
idx = (y * 80 + x) * 23 + 3
if (data[idx:idx + 5] < 8000).all():
dist[y, x, :] = 0
else:
dist[y, x, :] = data[idx + 15:idx + 20]
self.model = dist
def undistort(self, x, y, z):
i2 = int((z + 1) / 2)
i1 = i2 - 1
a = (z - (i1 * 2 + 1)) / 2
x = int(x / 8)
y = int(y / 6)
f = (1 - a) * self.model[y, x, min(max(i1, 0), 4)] + a * self.model[
y, x, min(i2, 4)]
if f == 0:
return 0
else:
return z / f
def simulate(self,
inputpng,
outputpng,
depth_scale=1000.0,
deterministic=False):
a = o3d.t.io.read_image(inputpng).as_tensor()
a = a.numpy().squeeze().astype(np.float32) / depth_scale
b = np.copy(a)
it = np.nditer(a, flags=['multi_index'], op_flags=['writeonly'])
while not it.finished:
# pixel shuffle
# here, 639 == width - 1, 479 == height - 1
if deterministic:
x = min(max(round(it.multi_index[1] + 0), 0), 639)
y = min(max(round(it.multi_index[0] + 0), 0), 479)
else:
x = min(
max(
round(it.multi_index[1] +
np.random.normal(0, 0.25)), 0), 639)
y = min(
max(
round(it.multi_index[0] +
np.random.normal(0, 0.25)), 0), 479)
# downsample
d = b[y - y % 2, x - x % 2]
# distortion
d = self.undistort(x, y, d)
# quantization and high freq noise
if d == 0:
it[0] = 0
else:
if deterministic:
it[0] = 35.130 * 8 / round((35.130 / d + 0) * 8)
else:
it[0] = 35.130 * 8 / round(
(35.130 / d + np.random.normal(0, 0.027778)) * 8)
it.iternext()
a = (a * depth_scale).astype(np.uint16)
a = np.expand_dims(a, axis=2)
o3d.t.io.write_image(outputpng, o3d.t.geometry.Image(a))
# Load dataset.
data = o3d.data.RedwoodIndoorLivingRoom1()
noise_model_path = data.noise_model_path
depth_scale = 1000.0
# Source image.
im_src_path = data.depth_paths[0]
im_src_uint16 = o3d.t.io.read_image(im_src_path).as_tensor().numpy()
im_src_float32 = im_src_uint16.astype(np.float32) / depth_scale
# Simulate "ground truth" noise depth image, with deterministic noise.
# We use the original simulator, which requires input and output path.
# See http://redwood-data.org/indoor/data/simdepth.py for the original
# implementation.
gt_simulator = Simulator()
gt_simulator.loaddistmodel(noise_model_path)
with tempfile.TemporaryDirectory() as dst_dir:
im_dst_path = os.path.join(dst_dir, "noisy_depth.png")
gt_simulator.simulate(im_src_path,
im_dst_path,
depth_scale=1000.0,
deterministic=True)
im_dst_gt = o3d.t.io.read_image(im_dst_path).as_tensor().numpy()
im_dst_gt = im_dst_gt.astype(np.float32) / depth_scale
# Our C++ implementation of the simulator.
simulator = o3d.t.io.DepthNoiseSimulator(noise_model_path)
simulator.enable_deterministic_debug_mode()
np.testing.assert_allclose(simulator.noise_model.numpy(),
gt_simulator.model)
# With uint16 input.
im_src = o3d.t.geometry.Image(im_src_uint16)
im_dst = simulator.simulate(im_src, depth_scale=depth_scale)
im_dst = im_dst.as_tensor().numpy().astype(np.float32) / depth_scale
np.testing.assert_allclose(im_dst, im_dst_gt)
# With float32 input.
im_src = o3d.t.geometry.Image(im_src_float32)
im_dst = simulator.simulate(im_src, depth_scale=1.0)
im_dst = im_dst.as_tensor().numpy()
im_dst = (im_dst * depth_scale).astype(np.uint16).astype(
np.float32) / depth_scale # Simulate rounding integers.
np.testing.assert_allclose(im_dst, im_dst_gt)
|
