-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathfind_outliers.py
207 lines (194 loc) · 9.54 KB
/
find_outliers.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
<
8000
div class="react-code-text react-code-line-contents" style="min-height:auto">
import random
import sys
import os
import argparse
import itertools
import csv
import numpy
import torch
import cv2
import torchvision
def parse_arguments(args):
desc = (
"3D60 dataset outliers generator."
)
parser = argparse.ArgumentParser(description=desc)
parser.add_argument('--action', type=str, \
default='calc', help='The action that will be run with \'calc\' used to find the outliers and \'save\' used to save them as tiled images.')
# paths
parser.add_argument("--suncg_path", type=str,\
default=argparse.SUPPRESS,\
help="Path to the rendered data of SunCG")
parser.add_argument("--s2d3d_path", type=str,\
default=argparse.SUPPRESS,\
help="Path to the rendered data of Stanford2D3D")
parser.add_argument("--m3d_path", type=str,\
default=argparse.SUPPRESS,\
help="Path to the rendered data of Matterport3D")
parser.add_argument("--outliers_path", type=str,\
default=".\\splits\\", \
help="The path where the generated outliers files will be saved at.")
# thresholds
parser.add_argument("--lower_threshold", type=float,\
default=0.5, help="Near (i.e. lower) distance value threshold.")
parser.add_argument("--upper_threshold", type=float,\
default=8.0, help="Far (i.e. upper) distance value threshold.")
parser.add_argument("--m3d_lower_bound", type=float,\
default=0.05, help="Near pixel percentage threshold for images that will be rejected as bad renders for Matterport3D.")
parser.add_argument("--m3d_upper_bound", type=float,\
default=0.25, help="Far pixel percentage threshold for images that will be rejected as bad renders for Matterport3D.")
parser.add_argument("--s2d3d_lower_bound", type=float,\
default=0.05, help="Near pixel percentage threshold for images that will be rejected as bad renders for Stanford2D3D.")
parser.add_argument("--s2d3d_upper_bound", type=float,\
default=0.2, help="Far pixel percentage threshold for images that will be rejected as bad renders for Stanford2D3D.")
parser.add_argument("--suncg_lower_bound", type=float,\
default=0.1, help="Near pixel percentage threshold for images that will be rejected as bad renders for SunCG.")
parser.add_argument("--suncg_upper_bound", type=float,\
default=0.2, help="Far pixel percentage threshold for images that will be rejected as bad renders for SunCG.")
return parser.parse_known_args(args)
def load_image(filename, data_type=torch.float32):
color_img = numpy.array(cv2.imread(filename, cv2.IMREAD_ANYCOLOR))
h, w, c = color_img.shape
color_data = color_img.astype(numpy.float32).transpose(2, 0, 1)
return torch.from_numpy(
color_data.reshape(1, c, h, w)
).type(data_type) / 255.0
def load_depth(filename, data_type=torch.float32):
dtmp = numpy.array(cv2.imread(filename, cv2.IMREAD_ANYDEPTH))
depth = torch.from_numpy(dtmp).type(data_type)
return depth.reshape(1, 1, depth.shape[0], depth.shape[1])
def find_outliers(folder, lower_threshold, upper_threshold,
lower_bound, upper_bound):
outliers = []
depth_files = [f for f in os.listdir(folder) if ".exr" in f and "_depth_" in f]
for depth_file in depth_files:
filename = os.path.join(folder, depth_file)
depth = load_depth(filename)
b, c, h, w = depth.size()
over = torch.sum(depth > upper_threshold).float()
under = torch.sum(depth < lower_threshold).float()
over_perc = over / (h * w)
under_perc = under / (h * w)
if over_perc > upper_bound or under_perc > lower_bound:
outliers.append(depth_file)
return outliers
def create_m3d_outliers(m3d_path, lower_threshold, upper_threshold,
lower_bound, upper_bound):
result = {}
result['m3d'] = list()
result['m3d'].extend(find_outliers(m3d_path, lower_threshold, upper_threshold, \
lower_bound, upper_bound))
return result
def create_s2d3d_outliers(s2d3d_path, lower_threshold, upper_threshold,
lower_bound, upper_bound):
result = {}
result['s2d3d'] = list()
for area in os.listdir(s2d3d_path):
result['s2d3d'].extend(find_outliers(os.path.join(s2d3d_path, area), \
lower_threshold, upper_threshold, lower_bound, upper_bound))
return result
def create_suncg_outliers(suncg_path, lower_threshold, upper_threshold,
lower_bound, upper_bound):
result = {}
result['scg'] = list()
result['scg'].extend(find_outliers(suncg_path, lower_threshold, upper_threshold, \
lower_bound, upper_bound))
return result
def dump_outliers(splits, args):
for outlier in outliers:
for k, v in outlier.items():
with open(os.path.join(args.outliers_path, '{}_outliers.csv'.format(k)), mode='w') as csv_file:
outlier_writer = csv.writer(csv_file, delimiter=',', \
lineterminator='\n', quotechar='"', quoting=csv.QUOTE_MINIMAL)
outlier_writer.writerow(['Outliers'])
for o in v:
outlier_writer.writerow([o])
def save_outliers(args):
paths, outliers_files, names = [], [], []
if 'suncg_path' in args:
paths.append(args.suncg_path)
outliers_files.append(os.path.join(args.outliers_path, "scg_outliers.csv"))
names.append('suncg')
if 'm3d_path' in args:
paths.append(args.m3d_path)
outliers_files.append(os.path.join(args.outliers_path, "m3d_outliers.csv"))
names.append('m3d')
if 's2d3d_path' in args:
paths.append(os.path.join(args.s2d3d_path, 'area1'))
paths.append(os.path.join(args.s2d3d_path, 'area2'))
paths.append(os.path.join(args.s2d3d_path, 'area3'))
paths.append(os.path.join(args.s2d3d_path, 'area4'))
paths.append(os.path.join(args.s2d3d_path, 'area5a'))
paths.append(os.path.join(args.s2d3d_path, 'area5b'))
paths.append(os.path.join(args.s2d3d_path, 'area6'))
outliers_files.append(os.path.join(args.outliers_path, "s2d3d_outliers.csv"))
outliers_files.append(os.path.join(args.outliers_path, "s2d3d_outliers.csv"))
outliers_files.append(os.path.join(args.outliers_path, "s2d3d_outliers.csv"))
outliers_files.append(os.path.join(args.outliers_path, "s2d3d_outliers.csv"))
outliers_files.append(os.path.join(args.outliers_path, "s2d3d_outliers.csv"))
outliers_files.append(os.path.join(args.outliers_path, "s2d3d_outliers.csv"))
outliers_files.append(os.path.join(args.outliers_path, "s2d3d_outliers.csv"))
names.append('s2d3d_a1')
names.append('s2d3d_a2')
names.append('s2d3d_a3')
names.append('s2d3d_a4')
names.append('s2d3d_a5a')
names.append('s2d3d_a5b')
names.append('s2d3d_a6')
for o, n, p in zip(outliers_files, names, paths):
with open(o, mode='r') as csv_file:
csv_reader = csv.reader(csv_file)
line_count = 0
viz_count = 0
saved_images = 0
tiles_per_image = 64
tensor = torch.zeros([tiles_per_image, 3, 256, 512])
for row in csv_reader:
if line_count == 0:
line_count += 1
else:
filename = os.path.join(p, row[0].replace('depth', 'color').replace('exr', 'png'))
if os.path.exists(filename) and '90.0' not in filename \
and '180.0' not in filename and '270.0' not in filename \
and 'Up' not in filename and 'Right' not in filename:
img = load_image(filename)
viz_count += 1
tensor[viz_count % tiles_per_image, :, :, :] = img
line_count += 1
if viz_count % tiles_per_image == 0 and viz_count != 0:
torchvision.utils.save_image(tensor, \
os.path.join(args.outliers_path, '{}_outliers_{}.png'.format(n, saved_images)))
tensor = torch.zeros([tiles_per_image, 3, 256, 512])
viz_count = 0
saved_images += 1
if viz_count < tiles_per_image:
torchvision.utils.save_image(tensor, \
os.path.join(args.outliers_path, '{}_outliers_{}.png'.format(n, saved_images)))
print(f'Processed {line_count} lines for {n}.')
if __name__ == "__main__":
args, unknown = parse_arguments(sys.argv)
if args.action == 'calc':
outliers = list()
''' Matterport3D '''
if 'm3d_path' in args:
m3d_outliers = create_m3d_outliers(args.m3d_path, \
args.lower_threshold, args.upper_threshold, \
args.m3d_lower_bound, args.m3d_upper_bound)
outliers.append(m3d_outliers)
''' Stanford2D3D '''
if 's2d3d_path' in args:
s2d3d_outliers = create_s2d3d_outliers(args.s2d3d_path, \
args.lower_threshold, args.upper_threshold, \
args.s2d3d_lower_bound, args.s2d3d_upper_bound)
outliers.append(s2d3d_outliers)
''' SunCG '''
if 'suncg_path' in args:
suncg_outliers = create_suncg_outliers(args.suncg_path, \
args.lower_threshold, args.upper_threshold, \
args.suncg_lower_bound, args.suncg_upper_bound)
outliers.append(suncg_outliers)
dump_outliers(outliers, args)
elif args.action == 'save':
save_outliers(args)
else:
print("Erroneous action selection, can only be one of calc/save.")