-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataset.py
291 lines (234 loc) · 10.2 KB
/
dataset.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
from torch.utils.data import DataLoader
import torch
import cv2
import os
import numpy as np
from PIL import Image, ImageDraw
from torchvision import transforms
import random
import torchvision.transforms.functional as F
import math
from colour.io.luts.iridas_cube import read_LUT_IridasCube
import logging
import glob
def augment(lut, input_image):
im_array = np.asarray(input_image, dtype=np.float32) / 255
is_non_default_domain = not np.array_equal(lut.domain, np.array([[0., 0., 0.], [1., 1., 1.]]))
dom_scale = None
if is_non_default_domain:
dom_scale = lut.domain[1] - lut.domain[0]
im_array = im_array * dom_scale + lut.domain[0]
im_array = lut.apply(im_array)
if is_non_default_domain:
im_array = (im_array - lut.domain[0]) / dom_scale
im_array = im_array * 255
aug_im = np.uint8(im_array)
return aug_im
def prepare_mask_and_masked_image(mask):
mask = np.array(mask.convert("L"))
mask = mask.astype(np.float32) / 255.0
mask = mask[None]
mask[mask < 0.5] = 0
mask[mask >= 0.5] = 1
mask = torch.from_numpy(mask)
return mask
# generate random masks
def random_mask(im_shape, ratio=1, mask_full_image=False):
mask = Image.new("L", im_shape, 0)
draw = ImageDraw.Draw(mask)
size = (random.randint(int(im_shape[0] * 0.1), int(im_shape[0] * 0.5)),
random.randint(int(im_shape[0] * 0.1), int(im_shape[1] * 0.5)))
# use this to always mask the whole image
if mask_full_image:
size = (int(im_shape[0] * ratio), int(im_shape[1] * ratio))
limits = (im_shape[0] - size[0] // 2, im_shape[1] - size[1] // 2)
center = (random.randint(size[0] // 2, limits[0]), random.randint(size[1] // 2, limits[1]))
draw_type = random.randint(0, 1)
if draw_type == 0 or mask_full_image:
draw.rectangle(
(center[0] - size[0] // 2, center[1] - size[1] // 2, center[0] + size[0] // 2, center[1] + size[1] // 2),
fill=255,
)
else:
draw.ellipse(
(center[0] - size[0] // 2, center[1] - size[1] // 2, center[0] + size[0] // 2, center[1] + size[1] // 2),
fill=255,
)
return mask
def free_form_mask(resolution):
ratio = resolution // 256
min_num_vertex = 4 * ratio
max_num_vertex = 12 * ratio
mean_angle = 2 * math.pi / 5
angle_range = 2 * math.pi / 15
min_width = 12 * ratio
max_width = 40 * ratio
H = W = resolution
average_radius = math.sqrt(H * H + W * W) / 8
mask = Image.new('L', (W, H), 0)
for _ in range(np.random.randint(1, 4)):
num_vertex = np.random.randint(min_num_vertex, max_num_vertex)
angle_min = mean_angle - np.random.uniform(0, angle_range)
angle_max = mean_angle + np.random.uniform(0, angle_range)
angles = []
vertex = []
for i in range(num_vertex):
if i % 2 == 0:
angles.append(2 * math.pi - np.random.uniform(angle_min, angle_max))
else:
angles.append(np.random.uniform(angle_min, angle_max))
h, w = mask.size
vertex.append((int(np.random.randint(0, w)), int(np.random.randint(0, h))))
for i in range(num_vertex):
r = np.clip(
np.random.normal(loc=average_radius, scale=average_radius // 2),
0, 2 * average_radius)
new_x = np.clip(vertex[-1][0] + r * math.cos(angles[i]), 0, w)
new_y = np.clip(vertex[-1][1] + r * math.sin(angles[i]), 0, h)
vertex.append((int(new_x), int(new_y)))
draw = ImageDraw.Draw(mask)
width = int(np.random.uniform(min_width, max_width))
draw.line(vertex, fill=255, width=width)
for v in vertex:
draw.ellipse((v[0] - width // 2,
v[1] - width // 2,
v[0] + width // 2,
v[1] + width // 2),
fill=255)
if np.random.normal() > 0:
mask.transpose(Image.FLIP_LEFT_RIGHT)
if np.random.normal() > 0:
mask.transpose(Image.FLIP_TOP_BOTTOM)
return mask
class Mask_Compose(transforms.Compose):
def __call__(self, image, mask=None):
for t in self.transforms:
if mask is None:
image = t(image)
else:
image, mask = t(image, mask)
if mask is None:
return image
else:
return image, mask
class Mask_Resize(transforms.Resize):
def forward(self, img, mask=None):
min_resolution = self.size
scale = random.uniform(0.5, 1.0)
self.target_size = [int(img.size[1] * scale), int(img.size[0] * scale)]
# Ensure that the minimum resolution is larger than args.resolution, or will raise error in Mask_RandomCrop().
if min(self.target_size) < min_resolution:
scale = min_resolution / min(self.target_size)
self.target_size = [int(self.target_size[0] * scale + 0.5), int(self.target_size[1] * scale + 0.5)]
img = F.resize(img, self.target_size, self.interpolation, self.max_size, self.antialias)
if mask is not None:
mask = F.resize(mask, self.target_size, F.InterpolationMode.NEAREST, self.max_size, self.antialias)
return img, mask
else:
return img
class Mask_RandomCrop(transforms.RandomCrop):
def forward(self, img, mask=None):
i, j, h, w = self.get_params(img, self.size)
if mask is not None:
return F.crop(img, i, j, h, w), F.crop(mask, i, j, h, w)
else:
return F.crop(img, i, j, h, w)
class Mask_CenterCrop(transforms.CenterCrop):
def forward(self, img, mask=None):
if mask is not None:
return F.center_crop(img, self.size), F.center_crop(mask, self.size)
else:
return F.center_crop(img, self.size)
class Mask_RandomHorizontalFlip(transforms.RandomHorizontalFlip):
def forward(self, img, mask=None):
if mask is None:
if torch.rand(1) < self.p:
return F.hflip(img)
return img
else:
if torch.rand(1) < self.p:
return F.hflip(img), F.hflip(mask)
return img, mask
# Pytorch dataset preparation.
class dataset_generation(torch.utils.data.Dataset):
def __init__(self, args):
self.args = args
assert args.image_path is not None or args.image_root is not None, \
'Either image_path or image_root should be given.'
self.image_files = [args.image_path]
if args.image_root is not None:
self.image_files = sorted(glob.glob(os.path.join(args.image_root, '*')))
self.pre_transforms = Mask_Compose(
[
Mask_Resize(args.resolution, interpolation=transforms.InterpolationMode.BILINEAR),
Mask_RandomCrop(args.resolution),
Mask_RandomHorizontalFlip(),
]
)
self.train_transforms = transforms.Compose(
[
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5]),
]
)
self.luts = []
file_paths = os.walk(os.path.join(args.lut_path, ''))
for root, dirs, files in file_paths:
for f in files:
file_path = os.path.join(root, f)
if not os.path.isfile(file_path):
continue
elif not file_path.lower().endswith('.cube'):
continue
else:
self.luts.append(read_LUT_IridasCube(file_path))
print('Loaded %d LUTs.' % len(self.luts))
logging.info('Loaded %d LUTs.' % len(self.luts))
def __getitem__(self, _):
img_path = random.choice(self.image_files)
self.image = cv2.cvtColor(cv2.imread(img_path), cv2.COLOR_BGR2RGB)
img = self.image.copy()
img = Image.fromarray(img)
crop_gt = self.pre_transforms(img)
mask_generate_type = random.randint(0, 1)
if mask_generate_type == 0:
# generate a random mask
crop_mask = random_mask(crop_gt.size, 1, False)
# prepare mask and masked image
crop_mask = prepare_mask_and_masked_image(crop_mask)
else:
# use the classic free-form masks in inpainting field
crop_mask = free_form_mask(crop_gt.size[0])
crop_mask = prepare_mask_and_masked_image(crop_mask)
img_resize = transforms.Resize([self.args.resolution, self.args.resolution])(Image.fromarray(self.image))
# Whether to apply LUT to both Foreground and Background.
# The threshold is set to not excessively change the color of the background.
is_dual_lut = (random.uniform(0.0, 1.0) > 0.8)
if is_dual_lut:
fore_lut, back_lut = random.sample(self.luts, 2)
crop_gt = np.array(crop_gt)
crop_mask = np.array(crop_mask)
img_resize = np.array(img_resize)
img_resize = augment(back_lut, img_resize)
crop_gt_origin = crop_gt.copy()
crop_gt = augment(back_lut, crop_gt)
crop_aug = augment(fore_lut, crop_gt_origin)
crop_aug = crop_aug * np.transpose(crop_mask, (1, 2, 0)) + crop_gt * (
1 - np.transpose(crop_mask, (1, 2, 0)))
crop_aug = crop_aug.astype(np.uint8)
LUT_reverse = fore_lut.invert(size=self.args.LUT_dim)
crop_aug_reverse = augment(LUT_reverse, crop_aug)
else:
lut = random.sample(self.luts, 1)[0]
crop_gt = np.array(crop_gt)
crop_mask = np.array(crop_mask)
crop_aug = augment(lut, crop_gt)
crop_aug = crop_aug * np.transpose(crop_mask, (1, 2, 0)) + crop_gt * (
1 - np.transpose(crop_mask, (1, 2, 0)))
crop_aug = crop_aug.astype(np.uint8)
LUT_reverse = lut.invert(size=self.args.LUT_dim)
crop_aug_reverse = augment(LUT_reverse, crop_aug)
return self.train_transforms(img_resize), self.train_transforms(crop_aug), crop_mask, self.train_transforms(
crop_gt), LUT_reverse.table, self.train_transforms(crop_aug_reverse), is_dual_lut
def __len__(self):
return int(self.args.iterations * self.args.batch_size)