forked from xavysp/DexiNed
-
Notifications
You must be signed in to change notification settings - Fork 0
/
datasets.py
486 lines (438 loc) · 18 KB
/
datasets.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
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
import os
import random
import cv2
import numpy as np
import torch
from torch.utils.data import Dataset
import json
DATASET_NAMES = [
'BIPED',
'BSDS',
'BSDS-RIND',
'BSDS300',
'CID',
'DCD',
'MDBD', #5
'PASCAL',
'NYUD',
'CLASSIC'
] # 8
def dataset_info(dataset_name, is_linux=True):
if is_linux:
config = {
'BSDS': {
'img_height': 512, #321
'img_width': 512, #481
'train_list': 'train_pair.lst',
'test_list': 'test_pair.lst',
'data_dir': '/opt/dataset/BSDS', # mean_rgb
'yita': 0.5
},
'BSDS-RIND': {
'img_height': 512, # 321
'img_width': 512, # 481
'train_list': 'train_pair2.lst',
'test_list': 'test_pair.lst',
'data_dir': '/opt/dataset/BSDS-RIND', # mean_rgb
'yita': 0.5
},
'BSDS300': {
'img_height': 512, #321
'img_width': 512, #481
'test_list': 'test_pair.lst',
'train_list': None,
'data_dir': '/opt/dataset/BSDS300', # NIR
'yita': 0.5
},
'PASCAL': {
'img_height': 416, # 375
'img_width': 512, #500
'test_list': 'test_pair.lst',
'train_list': None,
'data_dir': '/opt/dataset/PASCAL', # mean_rgb
'yita': 0.3
},
'CID': {
'img_height': 512,
'img_width': 512,
'test_list': 'test_pair.lst',
'train_list': None,
'data_dir': '/opt/dataset/CID', # mean_rgb
'yita': 0.3
},
'NYUD': {
'img_height': 448,#425
'img_width': 560,#560
'test_list': 'test_pair.lst',
'train_list': None,
'data_dir': '/opt/dataset/NYUD', # mean_rgb
'yita': 0.5
},
'MDBD': {
'img_height': 720,
'img_width': 1280,
'test_list': 'test_pair.lst',
'train_list': 'train_pair.lst',
'data_dir': '/opt/dataset/MDBD', # mean_rgb
'yita': 0.3
},
'BIPED': {
'img_height': 720, #720 # 1088
'img_width': 1280, # 1280 5 1920
'test_list': 'test_pair.lst',
'train_list': 'train_rgb.lst',
'data_dir': '/opt/dataset/BIPED', # mean_rgb
'yita': 0.5
},
'CLASSIC': {
'img_height': 512,
'img_width': 512,
'test_list': None,
'train_list': None,
'data_dir': 'data', # mean_rgb
'yita': 0.5
},
'DCD': {
'img_height': 352, #240
'img_width': 480,# 360
'test_list': 'test_pair.lst',
'train_list': None,
'data_dir': '/opt/dataset/DCD', # mean_rgb
'yita': 0.2
}
}
else:
config = {
'BSDS': {'img_height': 512, # 321
'img_width': 512, # 481
'test_list': 'test_pair.lst',
'data_dir': 'C:/Users/xavysp/dataset/BSDS', # mean_rgb
'yita': 0.5},
'BSDS300': {'img_height': 512, # 321
'img_width': 512, # 481
'test_list': 'test_pair.lst',
'data_dir': 'C:/Users/xavysp/dataset/BSDS300', # NIR
'yita': 0.5},
'PASCAL': {'img_height': 375,
'img_width': 500,
'test_list': 'test_pair.lst',
'data_dir': 'C:/Users/xavysp/dataset/PASCAL', # mean_rgb
'yita': 0.3},
'CID': {'img_height': 512,
'img_width': 512,
'test_list': 'test_pair.lst',
'data_dir': 'C:/Users/xavysp/dataset/CID', # mean_rgb
'yita': 0.3},
'NYUD': {'img_height': 425,
'img_width': 560,
'test_list': 'test_pair.lst',
'data_dir': 'C:/Users/xavysp/dataset/NYUD', # mean_rgb
'yita': 0.5},
'MDBD': {'img_height': 720,
'img_width': 1280,
'test_list': 'test_pair.lst',
'train_list': 'train_pair.lst',
'data_dir': 'C:/Users/xavysp/dataset/MDBD', # mean_rgb
'yita': 0.3},
'BIPED': {'img_height': 720, # 720
'img_width': 1280, # 1280
'test_list': 'test_pair.lst',
'train_list': 'train_rgb.lst',
'data_dir': 'C:/Users/xavysp/dataset/BIPED', # WIN: '../.../dataset/BIPED/edges'
'yita': 0.5},
'CLASSIC': {'img_height': 512,
'img_width': 512,
'test_list': None,
'train_list': None,
'data_dir': 'data', # mean_rgb
'yita': 0.5},
'DCD': {'img_height': 240,
'img_width': 360,
'test_list': 'test_pair.lst',
'data_dir': 'C:/Users/xavysp/dataset/DCD', # mean_rgb
'yita': 0.2}
}
return config[dataset_name]
class TestDataset(Dataset):
def __init__(self,
data_root,
test_data,
mean_bgr,
img_height,
img_width,
test_list=None,
arg=None
):
if test_data not in DATASET_NAMES:
raise ValueError(f"Unsupported dataset: {test_data}")
self.data_root = data_root
self.test_data = test_data
self.test_list = test_list
self.args=arg
# self.arg = arg
# self.mean_bgr = arg.mean_pixel_values[0:3] if len(arg.mean_pixel_values) == 4 \
# else arg.mean_pixel_values
self.mean_bgr = mean_bgr
self.img_height = img_height
self.img_width = img_width
self.data_index = self._build_index()
print(f"mean_bgr: {self.mean_bgr}")
def _build_index(self):
sample_indices = []
if self.test_data == "CLASSIC":
# for single image testing
images_path = os.listdir(self.data_root)
labels_path = None
sample_indices = [images_path, labels_path]
else:
# image and label paths are located in a list file
if not self.test_list:
raise ValueError(
f"Test list not provided for dataset: {self.test_data}")
list_name = os.path.join(self.data_root, self.test_list)
if self.test_data.upper()=='BIPED':
with open(list_name) as f:
files = json.load(f)
for pair in files:
tmp_img = pair[0]
tmp_gt = pair[1]
sample_indices.append(
(os.path.join(self.data_root, tmp_img),
os.path.join(self.data_root, tmp_gt),))
else:
with open(list_name, 'r') as f:
files = f.readlines()
files = [line.strip() for line in files]
pairs = [line.split() for line in files]
for pair in pairs:
tmp_img = pair[0]
tmp_gt = pair[1]
sample_indices.append(
(os.path.join(self.data_root, tmp_img),
os.path.join(self.data_root, tmp_gt),))
return sample_indices
def __len__(self):
return len(self.data_index[0]) if self.test_data.upper()=='CLASSIC' else len(self.data_index)
def __getitem__(self, idx):
# get data sample
# image_path, label_path = self.data_index[idx]
if self.data_index[1] is None:
image_path = self.data_index[0][idx]
else:
image_path = self.data_index[idx][0]
label_path = None if self.test_data == "CLASSIC" else self.data_index[idx][1]
img_name = os.path.basename(image_path)
file_name = os.path.splitext(img_name)[0] + ".png"
# base dir
if self.test_data.upper() == 'BIPED':
img_dir = os.path.join(self.data_root, 'imgs', 'test')
gt_dir = os.path.join(self.data_root, 'edge_maps', 'test')
elif self.test_data.upper() == 'CLASSIC':
img_dir = self.data_root
gt_dir = None
else:
img_dir = self.data_root
gt_dir = self.data_root
# load data
image = cv2.imread(os.path.join(img_dir, image_path), cv2.IMREAD_COLOR)
if not self.test_data == "CLASSIC":
label = cv2.imread(os.path.join(
gt_dir, label_path), cv2.IMREAD_COLOR)
else:
label = None
im_shape = [image.shape[0], image.shape[1]]
image, label = self.transform(img=image, gt=label)
return dict(images=image, labels=label, file_names=file_name, image_shape=im_shape)
def transform(self, img, gt):
# gt[gt< 51] = 0 # test without gt discrimination
if self.test_data == "CLASSIC":
img_height = self.img_height
img_width = self.img_width
print(
f"actual size: {img.shape}, target size: {( img_height,img_width,)}")
# img = cv2.resize(img, (self.img_width, self.img_height))
img = cv2.resize(img, (img_width,img_height))
gt = None
# Make images and labels at least 512 by 512
elif img.shape[0] < 512 or img.shape[1] < 512:
img = cv2.resize(img, (self.args.test_img_width, self.args.test_img_height)) # 512
gt = cv2.resize(gt, (self.args.test_img_width, self.args.test_img_height)) # 512
# Make sure images and labels are divisible by 2^4=16
elif img.shape[0] % 16 != 0 or img.shape[1] % 16 != 0:
img_width = ((img.shape[1] // 16) + 1) * 16
img_height = ((img.shape[0] // 16) + 1) * 16
img = cv2.resize(img, (img_width, img_height))
gt = cv2.resize(gt, (img_width, img_height))
else:
img_width =self.args.test_img_width
img_height =self.args.test_img_height
img = cv2.resize(img, (img_width, img_height))
gt = cv2.resize(gt, (img_width, img_height))
# if self.yita is not None:
# gt[gt >= self.yita] = 1
img = np.array(img, dtype=np.float32)
# if self.rgb:
# img = img[:, :, ::-1] # RGB->BGR
img -= self.mean_bgr
img = img.transpose((2, 0, 1))
img = torch.from_numpy(img.copy()).float()
if self.test_data == "CLASSIC":
gt = np.zeros((img.shape[:2]))
gt = torch.from_numpy(np.array([gt])).float()
else:
gt = np.array(gt, dtype=np.float32)
if len(gt.shape) == 3:
gt = gt[:, :, 0]
gt /= 255.
gt = torch.from_numpy(np.array([gt])).float()
return img, gt
class BipedDataset(Dataset):
train_modes = ['train', 'test', ]
dataset_types = ['rgbr', ]
data_types = ['aug', ]
def __init__(self,
data_root,
img_height,
img_width,
mean_bgr,
train_mode='train',
dataset_type='rgbr',
# is_scaling=None,
# Whether to crop image or otherwise resize image to match image height and width.
crop_img=False,
arg=None
):
self.data_root = data_root
self.train_mode = train_mode
self.dataset_type = dataset_type
self.data_type = 'aug' # be aware that this might change in the future
self.img_height = img_height
self.img_width = img_width
self.mean_bgr = mean_bgr
self.crop_img = crop_img
self.arg = arg
self.data_index = self._build_index()
def _build_index(self):
assert self.train_mode in self.train_modes, self.train_mode
assert self.dataset_type in self.dataset_types, self.dataset_type
assert self.data_type in self.data_types, self.data_type
data_root = os.path.abspath(self.data_root)
sample_indices = []
if self.arg.train_data.lower()=='biped':
images_path = os.path.join(data_root,
'edges/imgs',
self.train_mode,
self.dataset_type,
self.data_type)
labels_path = os.path.join(data_root,
'edges/edge_maps',
self.train_mode,
self.dataset_type,
self.data_type)
for directory_name in os.listdir(images_path):
image_directories = os.path.join(images_path, directory_name)
for file_name_ext in os.listdir(image_directories):
file_name = os.path.splitext(file_name_ext)[0]
sample_indices.append(
(os.path.join(images_path, directory_name, file_name + '.jpg'),
os.path.join(labels_path, directory_name, file_name + '.png'),)
)
else:
file_path = os.path.join(data_root, self.arg.train_list)
if self.arg.train_data.lower()=='bsds':
with open(file_path, 'r') as f:
files = f.readlines()
files = [line.strip() for line in files]
pairs = [line.split() for line in files]
for pair in pairs:
tmp_img = pair[0]
tmp_gt = pair[1]
sample_indices.append(
(os.path.join(data_root,tmp_img),
os.path.join(data_root,tmp_gt),))
else:
with open(file_path) as f:
files = json.load(f)
for pair in files:
tmp_img = pair[0]
tmp_gt = pair[1]
sample_indices.append(
(os.path.join(data_root, tmp_img),
os.path.join(data_root, tmp_gt),))
return sample_indices
def __len__(self):
return len(self.data_index)
def __getitem__(self, idx):
# get data sample
image_path, label_path = self.data_index[idx]
# load data
image = cv2.imread(image_path, cv2.IMREAD_COLOR)
label = cv2.imread(label_path, cv2.IMREAD_GRAYSCALE)
image, label = self.transform(img=image, gt=label)
return dict(images=image, labels=label)
def transform(self, img, gt):
gt = np.array(gt, dtype=np.float32)
if len(gt.shape) == 3:
gt = gt[:, :, 0]
gt /= 255. # for DexiNed input and BDCN
img = np.array(img, dtype=np.float32)
img -= self.mean_bgr
i_h, i_w,_ = img.shape
# data = []
# if self.scale is not None:
# for scl in self.scale:
# img_scale = cv2.resize(img, None, fx=scl, fy=scl, interpolation=cv2.INTER_LINEAR)
# data.append(torch.from_numpy(img_scale.transpose((2, 0, 1))).float())
# return data, gt
# 400 for BIPEd and 352 for BSDS check with 384
crop_size = self.img_height if self.img_height == self.img_width else None#448# MDBD=480 BIPED=480/400 BSDS=352
# # for BSDS 352/BRIND
# if i_w> crop_size and i_h>crop_size:
# i = random.randint(0, i_h - crop_size)
# j = random.randint(0, i_w - crop_size)
# img = img[i:i + crop_size , j:j + crop_size ]
# gt = gt[i:i + crop_size , j:j + crop_size ]
# for BIPED/MDBD
if np.random.random() > 0.4: #l
h,w = gt.shape
if i_w> 500 and i_h>500:
LR_img_size = crop_size #l BIPED=256, 240 200 # MDBD= 352 BSDS= 176
i = random.randint(0, h - LR_img_size)
j = random.randint(0, w - LR_img_size)
# if img.
img = img[i:i + LR_img_size , j:j + LR_img_size ]
gt = gt[i:i + LR_img_size , j:j + LR_img_size ]
else:
LR_img_size = 352#256 # l BIPED=208-352, # MDBD= 352-480- BSDS= 176-320
i = random.randint(0, h - LR_img_size)
j = random.randint(0, w - LR_img_size)
# if img.
img = img[i:i + LR_img_size, j:j + LR_img_size]
gt = gt[i:i + LR_img_size, j:j + LR_img_size]
img = cv2.resize(img, dsize=(crop_size, crop_size), )
gt = cv2.resize(gt, dsize=(crop_size, crop_size))
else:
# New addidings
img = cv2.resize(img, dsize=(crop_size, crop_size))
gt = cv2.resize(gt, dsize=(crop_size, crop_size))
# BSDS
# gt[gt>0.28]=1. # BSDS/MDBD
# gt[gt<=0.28]=0. # BSDS/MDBD
# for BIPED / BRIND
gt[gt > 0.2] += 0.6# 0.5 for BIPED/BSDS-RIND
gt = np.clip(gt, 0., 1.) # BIPED/BSDS-RIND
# # for MDBD
# gt[gt > 0.1] +=0.7
# gt = np.clip(gt, 0., 1.)
# # For RCF input
# # -----------------------------------
# gt[gt==0]=0.
# gt[np.logical_and(gt>0.,gt<0.5)] = 2.
# gt[gt>=0.5]=1.
#
# gt = gt.astype('float32')
# ----------------------------------
img = img.transpose((2, 0, 1))
img = torch.from_numpy(img.copy()).float()
gt = torch.from_numpy(np.array([gt])).float()
return img, gt