-
Notifications
You must be signed in to change notification settings - Fork 0
/
preprocessing.py
165 lines (137 loc) · 5.97 KB
/
preprocessing.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
import os
import sys
import numpy as np
import cv2
import imageprocessing as ip
import argparse
from tqdm import tqdm
import scipy.io as sio
import cv2
MODELS = ['retinaface', 'tinaface', 'dsfd']
def generateNoiseImage(input_path, output_path, noise, level):
img = cv2.imread(input_path)
assert type(img) != type(None)
#img_name = input_path.split('/')[-1]
if noise == 'gaussian_noise':
_img = ip.gaussian(img, level)
elif noise == 'salt_pepper':
_img = ip.salt_pepper(img, level)
elif noise == 'poisson':
_img = ip.poisson(img, level)
elif noise == 'speckle':
_img = ip.speckle(img, 0.3, level)
elif noise == 'gamma':
_img = ip.gamma(img, level)
else:
raise Exception('Specified noise invalid!')
cv2.imwrite(output_path, _img)
def WIDER_procedure(noise, level, output_path, mat_file, original_path):
EVENTS = mat_file['event_list']
FILE_LIST = mat_file['file_list']
for i, event in enumerate(tqdm(EVENTS)):
_event = event[0][0]
os.mkdir(
os.path.join(output_path, 'images', _event)
)
for filename_list in FILE_LIST[i]:
for filename in filename_list:
_filename = filename[0][0]+'.jpg'
_in = os.path.join(original_path, 'WIDER_val/images', _event, _filename)
_out = os.path.join(output_path, 'images', _event, _filename)
generateNoiseImage(_in, _out, noise, level)
def parseArgs():
parser = argparse.ArgumentParser()
parser.add_argument('--type', type=str)
parser.add_argument('--data_root', type=str)
parser.add_argument('--noise_list', type=str)
parser.add_argument('--correction_list', type=str)
return parser.parse_args()
def readNoiseText(contents):
# content is a list of the file contents:
noise_dict = {}
for con in contents:
noise, params = con.split(' ')
param = params.split(',')
param = list(map(float, param))
noise_dict[noise] = param
return noise_dict
def generateValList(parent_dir, wider_val_mat):
EVENTS = wider_val_mat['event_list']
FILE_LIST = wider_val_mat['file_list']
with open(os.path.join(parent_dir, 'val.txt'), 'w') as f:
for i, event in enumerate(EVENTS):
for filename_list in FILE_LIST[i]:
for filename in filename_list:
_event = event[0][0]
_filename = filename[0][0]+'.jpg'
_out = os.path.join(parent_dir, 'images', _event, _filename)
f.write('%s\n'%_out)
if __name__ == '__main__':
args = parseArgs()
wider_val_mat = sio.loadmat('./WIDERFACE/eval_tools/ground_truth/wider_face_val.mat')
ROOT = os.getcwd()
print(ROOT)
# make sure that a folder in the parent level created:
ROOT_NOISE = 'NOISES'
CORRECT_ROOT = 'CORRECTIONS'
EVENTS = wider_val_mat['event_list']
if not os.path.exists(ROOT_NOISE):
os.mkdir(ROOT_NOISE)
if not os.path.exists(CORRECT_ROOT):
os.mkdir(CORRECT_ROOT)
# parse out the noises you need to run:
with open(args.noise_list, 'r') as f:
content = f.readlines()
content = list(map(str.strip, content))
noises = readNoiseText(content)
# do the same for corrections:
with open(args.correction_list,'r') as f:
content = f.readlines()
content = list(map(str.strip, content))
corrections = readNoiseText(content)
print("Now running following noises: ", noises.keys())
for item in noises.items():
print("Running: ", item[0])
NOISE_DIR = os.path.join(ROOT_NOISE, "%s"%(item[0]) )
if not os.path.exists( NOISE_DIR ):
os.mkdir(NOISE_DIR)
for param in tqdm(item[1]):
SUB_NOISE_DIR = os.path.join(NOISE_DIR, '%0.6f'%(param))
if not os.path.exists(SUB_NOISE_DIR):
os.mkdir(SUB_NOISE_DIR)
os.mkdir(os.path.join(SUB_NOISE_DIR, 'detections'))
os.mkdir(os.path.join(SUB_NOISE_DIR, 'images'))
# write out ./SUB_NOISE_DIR/detections/[model] folders:
for model in MODELS:
_model_path = os.path.join(SUB_NOISE_DIR, 'detections', model)
if not os.path.exists(_model_path):
os.mkdir(_model_path)
for event in EVENTS:
class_model = os.path.join(_model_path, event[0][0])
if not os.path.exists(class_model):
os.mkdir(class_model)
FULL_SUB_PATH = os.path.join(ROOT, SUB_NOISE_DIR)
# write out all correction folders needed:
for correct in corrections.items():
CORRECT_PATH = os.path.join(CORRECT_ROOT, '%s_%s'%(item[0], correct[0]))
if not os.path.exists(CORRECT_PATH):
os.mkdir(CORRECT_PATH)
for correct_param in correct[1]:
CORRECT_PARAM_PATH = os.path.join(CORRECT_PATH, "%0.6f_%0.6f"%(param, correct_param))
if not os.path.exists(CORRECT_PARAM_PATH):
os.mkdir(CORRECT_PARAM_PATH)
# write out ./CORRECTIONS/[noise_correction]/[model] folders:
for model in MODELS:
_model_path = os.path.join(CORRECT_PARAM_PATH, model)
if not os.path.exists(_model_path):
os.mkdir(_model_path)
for event in EVENTS:
class_model = os.path.join(_model_path, event[0][0])
if not os.path.exists(class_model):
os.mkdir(class_model)
#create if val file exists:
if not os.path.exists(
os.path.join(FULL_SUB_PATH, 'val.txt')
):
generateValList(FULL_SUB_PATH, wider_val_mat)
WIDER_procedure(item[0], param, SUB_NOISE_DIR, wider_val_mat, args.data_root)