-
Notifications
You must be signed in to change notification settings - Fork 68
/
data_augmentation.py
267 lines (227 loc) · 9.07 KB
/
data_augmentation.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
265
266
267
#
# Authors: Bowen Wen
# Contact: wenbowenxjtu@gmail.com
# Created in 2020
#
# Copyright (c) Rutgers University, 2020 All rights reserved.
#
# Wen, B., C. Mitash, B. Ren, and K. E. Bekris. "se (3)-TrackNet:
# Data-driven 6D Pose Tracking by Calibrating Image Residuals in
# Synthetic Domains." In IEEE/RSJ International Conference on Intelligent
# Robots and Systems (IROS). 2020.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# * Neither the name of the PRACSYS, Bowen Wen, Rutgers University,
# nor the names of its contributors may be used to
# endorse or promote products derived from this software without
# specific prior written permission.
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 'AS IS' AND
# ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
# WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE FOR ANY
# DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
# (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
# LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
# ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
# SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#
import scipy.signal
import scipy.stats
import torch
import sys,os,time
dir_path = os.path.dirname(os.path.realpath(__file__))
from PIL import Image
import numpy as np
import cv2
from Utils import *
class HSVJitter(object):
def __init__(self, h_noise, s_noise, v_noise, prob=0.5):
self.prob = prob
self.h_noise = h_noise
self.s_noise = s_noise
self.v_noise = v_noise
def __call__(self, data):
rgbA, depthA, rgbB, depthB, maskA, maskB, poseA = data
mask = depthB>100
hsv = cv2.cvtColor(rgbB, cv2.COLOR_RGB2HSV).astype(np.float32)
H = hsv.shape[0]
W = hsv.shape[1]
if np.random.uniform() < self.prob:
hsv[:, :, 0] += np.random.uniform(-self.h_noise, self.h_noise)
if np.random.uniform() < self.prob:
hsv[:, :, 1] += np.random.uniform(-self.s_noise, self.s_noise)
if np.random.uniform() < self.prob:
hsv[:, :, 2] += np.random.uniform(-self.v_noise, self.v_noise)
hsv = np.clip(hsv,0,255)
rgbB[mask] = cv2.cvtColor(hsv.astype(np.uint8), cv2.COLOR_HSV2RGB)[mask]
rgbB = rgbB.astype(np.uint8)
return rgbA, depthA, rgbB, depthB, maskA, maskB,poseA
class ChangeBright:
def __init__(self,prob=0.5,mag=[0.5,1.5]):
self.mag = mag
def __call__(self,data):
rgbA, depthA, rgbB, depthB, maskA, maskB, poseA = data
rgbB = rgbB*random.uniform(self.mag[0],self.mag[1])
rgbB = np.clip(rgbB,0,255).astype(np.uint8)
return rgbA, depthA, rgbB, depthB, maskA, maskB,poseA
class GaussianNoise(object):
def __init__(self, rgb_noise, depth_noise, prob=0.5):
self.rgb_noise = rgb_noise
self.depth_noise = depth_noise
self.prob = prob
def __call__(self, data):
rgbA, depthA, rgbB, depthB,maskA, maskB, poseA = data
mask = depthB>100
if np.random.uniform() < self.prob:
std = np.random.uniform(0, self.rgb_noise)
noise = np.random.normal(0, std, size=rgbA.shape)
rgbB[mask] = rgbB[mask] + noise[mask]
if np.random.uniform() < self.prob:
std = np.random.uniform(0, self.depth_noise)
noise = np.random.normal(0, std, size=depthB.shape)
depthB[mask] = depthB[mask] + noise[mask]
return rgbA, depthA, rgbB, depthB, maskA, maskB,poseA
class GaussianBlur(object):
def __init__(self, max_kernel_size, min_kernel_size=3, prob=0.4):
self.prob = prob
self.max_kernel_size = max_kernel_size
self.min_kernel_size = 3
def __call__(self, data):
rgbA, depthA, rgbB, depthB, maskA, maskB,poseA = data
if np.random.uniform() < self.prob:
ksize = np.random.randint(1, self.max_kernel_size//2+1)
ksize = 2*ksize+1
rgbB = cv2.GaussianBlur(rgbB, (ksize,ksize), sigmaX=2)
if np.random.uniform() < self.prob:
ksize = np.random.randint(1, self.max_kernel_size//2+1)
ksize = 2*ksize+1
depthB = cv2.GaussianBlur(depthB, (ksize,ksize), sigmaX=2)
return rgbA, depthA, rgbB, depthB, maskA, maskB,poseA
class OffsetDepth(object):
def __init__(self):
pass
def __call__(self, data):
rgbA, depthA, rgbB, depthB, maskA, maskB,poseA = data
depthA = self.normalize_depth(depthA, poseA)
depthB = self.normalize_depth(depthB, poseA)
return rgbA.astype(np.float32), depthA, rgbB.astype(np.float32), depthB, maskA, maskB, poseA
def normalize_depth(self, depth, pose):
depth = depth.astype(np.float32)
invalid_mask = np.logical_or(depth<=100, depth>=2000)
if pose[2, 3]<0: #gl pose
depth += pose[2, 3] * 1000
else:
depth -= pose[2, 3] * 1000
depth[invalid_mask] = 2000
assert (depth<=2000).all()
return depth
class NormalizeChannels(object):
def __init__(self,mean,std):
self.mean = mean
self.std = std
def __call__(self, data):
rgbA, depthA, rgbB, depthB, maskA, maskB, poseA = data
rgbA, depthA = self.normalize_channels(rgbA, depthA, self.mean[:4], self.std[:4])
rgbB, depthB = self.normalize_channels(rgbB, depthB, self.mean[4:], self.std[4:])
return rgbA, depthA, rgbB, depthB, maskA, maskB,poseA
def normalize_channels(self, rgb, depth, mean, std):
rgb = rgb.transpose(2,0,1)
rgb = (rgb-mean[:3, np.newaxis, np.newaxis])/std[:3, np.newaxis, np.newaxis]
depth = (depth-mean[3, np.newaxis, np.newaxis])/std[3, np.newaxis, np.newaxis]
return rgb, depth
class Transpose(object):
def __call__(self, data):
rgbA, depthA, rgbB, depthB,maskA, maskB, poseA = data
rgbA = rgbA.transpose(2,0,1)
rgbB = rgbB.transpose(2,0,1)
return rgbA, depthA, rgbB, depthB, maskA, maskB,poseA
class ToTensor(object):
def __init__(self):
pass
def __call__(self, data):
rgbA, depthA, rgbB, depthB, maskA, maskB, poseA = data
bufferA = np.zeros((4, rgbA.shape[1], rgbA.shape[2]), dtype=np.float32)
bufferA[0:3, :, :] = rgbA
bufferA[3, :, :] = depthA
bufferB = np.zeros((4, rgbA.shape[1], rgbA.shape[2]), dtype=np.float32)
bufferB[0:3, :, :] = rgbB
bufferB[3, :, :] = depthB
bufferA = torch.from_numpy(bufferA)
bufferB = torch.from_numpy(bufferB)
return [bufferA, bufferB], maskA, maskB
def to_tensor(self, rgb,depth):
buffer = np.zeros((4, rgb.shape[1], rgb.shape[2]), dtype=np.float32)
buffer[0:3, :, :] = rgb
buffer[3, :, :] = depth
buffer = torch.from_numpy(buffer).float()
return buffer
class DepthMissing():
def __init__(self,prob=0.5,missing_percent=0.5):
self.prob = prob
self.missing_percent = missing_percent
def __call__(self,data):
rgbA, depthA, rgbB, depthB,maskA, maskB, poseA = data
W = depthB.shape[1]
H = depthB.shape[0]
us,vs = np.where(depthB>100)
if np.random.uniform(0,1)<self.prob:
missing_percent = np.random.uniform(0,self.missing_percent)
missing_ids = np.random.choice(np.arange(0,len(us)), int(missing_percent*len(us)), replace=False)
depthB[vs[missing_ids], us[missing_ids]] = 0
return rgbA, depthA, rgbB, depthB, maskA, maskB,poseA
class BlackCover():
'''Random black cover to imitate cases of object outside of image.
'''
def __init__(self, prob=0.3):
self.prob = prob
def __call__(self,data):
rgbA, depthA, rgbB, depthB, maskA, maskB,prior = data
rgbB_backup = rgbB.copy()
depthB_backup = depthB.copy()
maskB_backup = maskB.copy().astype(np.uint8)
num_valid = np.sum(maskB_backup)
if np.random.uniform(0,1) >= self.prob:
return rgbA, depthA, rgbB, depthB, maskA, maskB,prior
H = rgbB.shape[0]
W = rgbB.shape[1]
corner_uv = (np.random.randint(0,W),np.random.randint(0,H))
tlbr = np.random.choice([0,1,2,3])
i = tlbr
while True:
if i==0: #top left
rgbB[:corner_uv[1],:corner_uv[0],:] = 0
depthB[:corner_uv[1],:corner_uv[0]] = -9999
maskB[:corner_uv[1],:corner_uv[0]] = 0
elif i==1: #top right
rgbB[:corner_uv[1],corner_uv[0]:,:] = 0
depthB[:corner_uv[1],corner_uv[0]:] = -9999
maskB[:corner_uv[1],corner_uv[0]:] = 0
elif i==2: #bottom left
rgbB[corner_uv[1]:,:corner_uv[0],:] = 0
depthB[corner_uv[1]:,:corner_uv[0]] = -9999
maskB[corner_uv[1]:,:corner_uv[0]] = 0
elif i==3:
rgbB[corner_uv[1]:,corner_uv[0]:,:] = 0
depthB[corner_uv[1]:,corner_uv[0]:] = -9999
maskB[corner_uv[1]:,corner_uv[0]:] = 0
remainedB_valid = maskB==1
if np.sum(remainedB_valid)/float(num_valid)<0.5: # Make sure at least remain some visibility of object
rgbB = rgbB_backup.copy()
depthB = depthB_backup.copy()
maskB = maskB_backup.copy()
i += 1
i = i%4
else:
break
if i==tlbr:
corner_uv = (np.random.randint(0,W),np.random.randint(0,H))
tlbr = np.random.choice([0,1,2,3])
i = tlbr
return rgbA, depthA, rgbB, depthB, maskA, maskB,prior