-
Notifications
You must be signed in to change notification settings - Fork 12
/
prepare_data.py
322 lines (258 loc) · 15.9 KB
/
prepare_data.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
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
"""
Copyright (c) 2022 Samsung Electronics Co., Ltd.
Author(s):
Abhijith Punnappurath (abhijith.p@samsung.com)
Abdullah Abuolaim (abdullah.abuolaim@gmail.com)
Abdelrahman Abdelhamed (a.abdelhamed@samsung.com; abdoukamel@gmail.com)
Licensed under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) License, (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at https://creativecommons.org/licenses/by-nc/4.0
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.
For conditions of distribution and use, see the accompanying LICENSE.md file.
Description:
Generating synthetic night images from day images.
"""
import pickle
import numpy as np
import cv2
import os
import random
import argparse
from glob import glob
import scipy.io
from copy import deepcopy
from utils.relight import relight_locally, apply_local_lights_rgb
from pipeline.pipeline import run_pipeline
from pipeline.pipeline_utils import normalize, denormalize, get_visible_raw_image, ratios2floats, white_balance, \
get_metadata
from utils.gen_utils import check_dir
from noise_profiler.image_synthesizer import load_noise_model, synthesize_noisy_image_v2
noise_model_path = './noise_profiler/h-gauss-s20-v1'
noise_model, iso2b1_interp_splines, iso2b2_interp_splines = load_noise_model(path=noise_model_path)
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument('--base_address', type=str,
help='path to day dataset',
default='./dataset/day/'
)
parser.add_argument('--savefolderpath', default='synthetic_datasets', type=str, help='main path to save to')
parser.add_argument('--savefoldername', default='night', type=str, help='directory to save to')
parser.add_argument('--how_many_train', default=60, type=int, help='how many training images')
parser.add_argument('--dim', default=False, action='store_true', help='dim or not')
parser.add_argument('--relight', default=False, action='store_true', help='relight or not')
parser.add_argument('--discard_black_level', default=False, type=bool,
help='whether to discard black-level subtraction or not')
parser.add_argument('--clip', default=True, type=bool, help='whether to clip or not')
parser.add_argument('--relight_local', default=False, action='store_true', help='locally relight or not')
parser.add_argument('--min_num_lights', default=5, type=int, help='min. number of local lights')
parser.add_argument('--max_num_lights', default=5, type=int, help='max. number of local lights')
parser.add_argument('--min_light_size', default=.5, type=int, help='min. local light size as percent of image dim.')
parser.add_argument('--max_light_size', default=1., type=int, help='max. local light size as percent of image dim.')
parser.add_argument('--save_light_masks', default=False, action='store_true', help='whether to save light masks, '
'disable for speedup')
parser.add_argument('--num_sat_lights', default=5, type=int, help='number of small saturated local lights')
parser.add_argument('--iso_list', default='1600,3200', type=str,
help='list of discrete ISOs to pick from when adding noise, separate with comma')
args = parser.parse_args()
print(args)
return args
def get_illum_normalized_by_g(illum_in_arr):
return illum_in_arr[:, 0] / illum_in_arr[:, 1], illum_in_arr[:, 1] / illum_in_arr[:, 1], illum_in_arr[:, 2] / illum_in_arr[:, 1]
def synth_night_imgs(in_img_or_path, in_day_meta_data, dim=True, relight=True, iso=50,
discard_black_level=False, _clip=True, relight_local=True,
min_num_lights=5, max_num_lights=5, min_light_size = 0.5, max_light_size = 1.0, num_sat_lights=5):
"""
Synthesizing nigh time images
:param in_img_or_path:
:param in_day_meta_data:
:param dim:
:param relight:
:param iso:
:param discard_black_level:
:param _clip:
:param relight_local: Whether to locally relight image.
:param min_num_lights: Minimum number of local illuminants, in case of local relighting.
:param max_num_lights: Maximum number of local illuminants, in case of local relighting.
:param min_light_size: Minimum local light size as percent of image dimension, in case of local relighting.
:param max_light_size: Maximum local light size as percent of image dimension, in case of local relighting.
:param num_sat_lights: Number of small saturated local lights, in case of local relighting.
Return synthetic day-to-night image.
"""
if type(in_img_or_path) == str:
image_path = in_img_or_path
# raw image data
in_raw_image = get_visible_raw_image(image_path)
else:
in_raw_image = in_img_or_path
meta_data_night = deepcopy(in_day_meta_data)
if dim:
sampled_bright = random.uniform(0.55, 0.9)
else:
sampled_bright = 1.0
# relighting
if relight_local:
# local illuminants
num_of_samples = np.random.randint(min_num_lights + num_sat_lights, max_num_lights + num_sat_lights + 1, 1)
if discard_black_level:
sampled_wb = (np.random.multivariate_normal(gt_illum_mean, gt_illum_cov, num_of_samples) * (
1023 - 64) + 64) / 1023
else:
sampled_wb = np.random.multivariate_normal(gt_illum_mean, gt_illum_cov, num_of_samples)
elif relight:
num_of_samples = 1
if discard_black_level:
sampled_wb = (np.random.multivariate_normal(gt_illum_mean, gt_illum_cov, num_of_samples)[0] * (
1023 - 64) + 64) / 1023
else:
sampled_wb = np.random.multivariate_normal(gt_illum_mean, gt_illum_cov, num_of_samples)[0]
else:
sampled_wb = ratios2floats(in_day_meta_data['as_shot_neutral'])
if discard_black_level:
in_raw_image = normalize(in_raw_image, 0, in_day_meta_data['white_level'], clip=_clip)
else:
in_raw_image = normalize(in_raw_image, in_day_meta_data['black_level'], in_day_meta_data['white_level'],
clip=_clip)
white_balanceed_img = white_balance(in_raw_image, ratios2floats(in_day_meta_data['as_shot_neutral']),
in_day_meta_data['cfa_pattern'], clip=_clip)
if relight_local:
day_illum = ratios2floats(in_day_meta_data['as_shot_neutral'])
meta_data_night['as_shot_neutral'] = day_illum
meta_data_night['day_illuminant'] = day_illum
meta_data_night['avg_night_illuminant'] = np.mean(sampled_wb, axis=0)
else:
meta_data_night['as_shot_neutral'] = sampled_wb
meta_data_night['iso'] = iso # sampled_iso
white_balanceed_img_dim = white_balanceed_img * sampled_bright
sampled_wb_inv = list(1 / np.asarray(sampled_wb))
# white imbalance
if relight_local:
white_imbalanceed_img_dim_night, local_lights_ = relight_locally(white_balanceed_img_dim, sampled_wb,
in_day_meta_data['cfa_pattern'], clip=_clip,
invert_wb=True,
min_light_size=min_light_size,
max_light_size=max_light_size,
num_sat_lights=num_sat_lights)
else:
white_imbalanceed_img_dim_night = white_balance(white_balanceed_img_dim, sampled_wb_inv,
in_day_meta_data['cfa_pattern'], clip=_clip)
local_lights_ = None
if discard_black_level:
white_imbalanceed_img_dim_night = (
denormalize(white_imbalanceed_img_dim_night, 0, meta_data_night['white_level'], clip=_clip)).astype(
np.uint16)
else:
white_imbalanceed_img_dim_night = (
denormalize(white_imbalanceed_img_dim_night, meta_data_night['black_level'], meta_data_night['white_level'],
clip=_clip)).astype(np.uint16)
if relight_local:
return white_imbalanceed_img_dim_night, meta_data_night, local_lights_
else:
return white_imbalanceed_img_dim_night, meta_data_night
if __name__ == "__main__":
args = parse_args()
# load nighttime illuminants
gt_illum = scipy.io.loadmat('utils/gray_card_illum_dict.mat')
gt_illum = gt_illum['night_dict']
gt_illum[:, 0], gt_illum[:, 1], gt_illum[:, 2] = get_illum_normalized_by_g(gt_illum)
gt_illum_mean = np.mean(gt_illum, 0)
gt_illum_cov = np.cov(np.transpose(gt_illum))
# read day images
base_address = args.base_address
outdoor_daytime_img_names = [os.path.basename(x) for x in sorted(glob(os.path.join(base_address, '*.dng')))]
# create save directories
savefolder = os.path.join(args.savefolderpath, args.savefoldername)
check_dir(args.savefolderpath)
check_dir(savefolder)
for fol in ['train', 'val']: check_dir(os.path.join(savefolder, fol))
for fol in ['train', 'val']:
for subfol in ['clean_raw', 'noisy_raw', 'clean', 'noisy', 'metadata_raw']:
check_dir(os.path.join(savefolder, fol, subfol))
if args.relight_local:
light_mask_dirname = 'masks'
else:
light_mask_dirname = None
if args.save_light_masks and args.relight_local:
for fol in ['train', 'val']: check_dir(os.path.join(savefolder, fol, light_mask_dirname))
# camera pipeline parameters
stages = ['raw', 'active_area_cropping', 'linearization', 'normal', 'white_balance',
'demosaic', 'xyz', 'srgb', 'fix_orient', 'gamma', 'tone']
params = {
'save_as': 'png', # options: 'jpg', 'png', 'tif', etc.
'white_balancer': 'default', # options: default, or self-defined module
'demosaicer': 'menon2007', # options: '' for simple interpolation,
# 'EA' for edge-aware,
# 'VNG' for variable number of gradients,
# 'menon2007' for Menon's algorithm
'tone_curve': 'simple-s-curve', # options: 'simple-s-curve', 'default', or self-defined module
}
iso_list = [int(item) for item in args.iso_list.split(',')]
# main loop
for example_num in range(len(outdoor_daytime_img_names)):
print('Processing image:', example_num + 1, 'of', len(outdoor_daytime_img_names))
example_img_bayer_org = get_visible_raw_image(os.path.join(base_address, outdoor_daytime_img_names[example_num]))
meta_data_org = get_metadata(os.path.join(base_address, outdoor_daytime_img_names[example_num]))
rand_iso = iso_list[np.random.randint(2)]
results_ = synth_night_imgs(example_img_bayer_org, meta_data_org, dim=args.dim,
relight=args.relight, iso=rand_iso,
discard_black_level=args.discard_black_level,
_clip=args.clip, relight_local=args.relight_local,
min_num_lights=args.min_num_lights,
max_num_lights=args.max_num_lights,
min_light_size=args.min_light_size,
max_light_size=args.max_light_size,
num_sat_lights=args.num_sat_lights
)
if args.relight_local:
example_night_synth, meta_data_night, local_lights = results_
else:
example_night_synth, meta_data_night = results_
local_lights = None
if example_num >= args.how_many_train:
curfolder = os.path.join(savefolder, 'val')
else:
curfolder = os.path.join(savefolder, 'train')
cv2.imwrite(curfolder + '/clean_raw/' + outdoor_daytime_img_names[example_num][:-4] + '.png',
example_night_synth.astype(np.uint16))
pickle.dump(meta_data_night,
open(curfolder + '/metadata_raw/' + outdoor_daytime_img_names[example_num][:-4] + '.p', "wb"))
as_shot_neutral = meta_data_night['as_shot_neutral'] # keep as_shot_neutral
if args.relight_local:
meta_data_night['as_shot_neutral'] = meta_data_night['avg_night_illuminant'] # modify as_shot_neutral
night_synth_srgb_avg = run_pipeline(example_night_synth, params=params, metadata=meta_data_night, stages=stages)
night_synth_srgb_avg = (night_synth_srgb_avg * 255).astype(np.uint8)
cv2.imwrite(curfolder + '/clean/' + outdoor_daytime_img_names[example_num][:-4] + '.png', night_synth_srgb_avg[:, :, [2, 1, 0]])
meta_data_night['as_shot_neutral'] = as_shot_neutral # restore as_shot_neutral
noisy_night_image = synthesize_noisy_image_v2(example_night_synth, model=noise_model,
dst_iso=meta_data_night['iso'], min_val=0,
max_val=1023,
iso2b1_interp_splines=iso2b1_interp_splines,
iso2b2_interp_splines=iso2b2_interp_splines)
noisy_night_image = noisy_night_image.astype(np.uint16)
as_shot_neutral = meta_data_night['as_shot_neutral'] # keep as_shot_neutral
if args.relight_local:
meta_data_night['as_shot_neutral'] = meta_data_night['avg_night_illuminant'] # modify as_shot_neutral
noisy_night_image_srgb = run_pipeline(noisy_night_image, params=params, metadata=meta_data_night, stages=stages)
noisy_night_image_srgb = (noisy_night_image_srgb * 255).astype(np.uint8)
meta_data_night['as_shot_neutral'] = as_shot_neutral # restore as_shot_neutral
cv2.imwrite(curfolder + '/noisy/' + outdoor_daytime_img_names[example_num][:-4] + '_iso_' + str(rand_iso).zfill(4) + '.png', noisy_night_image_srgb[:, :, [2, 1, 0]])
cv2.imwrite(curfolder + '/noisy_raw/' + outdoor_daytime_img_names[example_num][:-4] + '.png',
noisy_night_image)
# save local light masks
if args.relight_local and args.save_light_masks and local_lights is not None:
neutral_image = np.ones(night_synth_srgb_avg.shape, dtype=np.float32)
# save individual local light masks
for k, light in enumerate(local_lights[1:]):
neutral_image_relight = apply_local_lights_rgb(neutral_image, [local_lights[0], light], clip=True,
invert_wb=True)
image_name = '{}_mask_{}_loc_{}_{}.jpg'.format(outdoor_daytime_img_names[example_num][:-4], k,
light.location[0], light.location[1])
cv2.imwrite(os.path.join(curfolder, light_mask_dirname, image_name),
(neutral_image_relight ** 0.4545 * 255).astype(np.uint8))
# save combined light mask
neutral_image_relight = apply_local_lights_rgb(neutral_image, local_lights, clip=True, invert_wb=True)
image_name = '{}_mask_combined.jpg'.format(outdoor_daytime_img_names[example_num][:-4])
cv2.imwrite(os.path.join(curfolder, light_mask_dirname, image_name),
(neutral_image_relight ** 0.4545 * 255).astype(np.uint8))
print('Done!')