-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathdataset.py
188 lines (148 loc) · 6.37 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
import os
import random
import numpy as np
import torch
import torch.utils.data as data
import torchvision.transforms.functional as transF
from PIL import Image
import torchvision.transforms as transforms
from utils import make_init_field, load_filenames, warp_position_map
transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
]
)
flip_index = [0, 3, 2, 1, 7, 6, 5, 4, 11, 10, 9, 8, 12, 13, 16, 15, 14]
def make_dataset(args):
list_total = os.listdir(os.path.join(args.data_root, 'image'))
train_list = random.sample(list_total, len(list_total) // 2)
if args.mode == 'train':
return WCDataSet(args, train_list, 'train')
else:
test_list = list(set(list_total).difference(set(train_list)))
return WCDataSet(args, test_list, 'test')
def load_img(path):
return Image.open(path).convert('RGB')
def load_landmark(path):
return torch.from_numpy(np.loadtxt(path, delimiter='\t')).float()
def cal_field(src, dst):
field_y, field_x = warp_position_map(src, dst)
field = np.concatenate((field_y, field_x), axis=2)
field = torch.from_numpy(field).float()
return field
def random_horizonal_flip(img, landmark):
if random.random() < 0.5:
img_hflip = transF.hflip(img)
landmark[:, 0] = 256 - landmark[:, 0]
landmark_hflip = torch.zeros_like(landmark)
for i in range(landmark.shape[0]):
landmark_hflip[i] = landmark[flip_index[i]]
return img_hflip, landmark_hflip
else:
return img, landmark
def random_enlarge_landmark(landmark_mean, landmark, p=0.5):
if random.random() < p:
rate = random.random() / 5 + 1
else:
rate = 1
landmark = (landmark - landmark_mean) * (rate - 1) + landmark
return landmark, rate
def random_resize_crop(img, landmark, resize=288):
if random.random() < 0.5:
w, h = img.size
time = resize / w
img = img.resize((resize, resize), Image.BILINEAR)
x = random.random() * (time - 1) * w
y = random.random() * (time - 1) * h
img_crop = img.crop((x, y, x + w, y + h))
landmark_crop = landmark * time
landmark_crop[:, 0] = landmark_crop[:, 0] - x
landmark_crop[:, 1] = landmark_crop[:, 1] - y
return img_crop, landmark_crop
else:
return img, landmark
class WCDataSet(data.Dataset):
def __init__(self, args, name_list, mode='train', transform=transform):
super(WCDataSet, self).__init__()
self.data_root = args.data_root
self.name_list = name_list
self.mode = mode
self.enlarge = args.enlarge
self.hflip = args.hflip
self.resize_crop = args.resize_crop
self.same_id = args.same_id
self.transform = transform
self.const_map = make_init_field().squeeze(0)
self.p_dir = {name: load_filenames(self.data_root, name, 'image', 'P') for name in name_list}
self.c_dir = {name: load_filenames(self.data_root, name, 'image', 'C') for name in name_list}
self.num_p = sum([len(files) for files in self.p_dir.values()])
self.num_c = sum([len(files) for files in self.c_dir.values()])
print('load dataset over')
print('{} image: {}, {} caricature: {}'.format(self.mode, self.num_p, self.mode, self.num_c))
self.img_list = []
for name in name_list:
self.img_list += self.p_dir[name]
assert len(self.img_list) == self.num_p
if self.mode == 'train':
self.size = min(args.max_dataset_size, 100000)
self.landmark_mean = torch.zeros(17, 2)
for img_path in self.img_list:
landmark_path = img_path.replace('image', 'landmark').replace('.jpg', '.txt')
landmark = load_landmark(landmark_path)
self.landmark_mean += landmark
self.landmark_mean /= self.num_p
else:
self.size = self.num_p
def sample_pair(self, same_id=True):
name1 = random.choice(self.name_list)
img_p_path = random.choice(self.p_dir[name1])
landmark_p_path = img_p_path.replace('image', 'landmark').replace('.jpg', '.txt')
if same_id:
img_c_path = random.choice(self.c_dir[name1])
else:
name2 = random.choice(self.name_list)
img_c_path = random.choice(self.c_dir[name2])
landmark_c_path = img_c_path.replace('image', 'landmark').replace('.jpg', '.txt')
return img_p_path, img_c_path, landmark_p_path, landmark_c_path
def __getitem__(self, index):
if self.mode == 'train':
img_p_path, img_c_path, landmark_p_path, landmark_c_path = self.sample_pair(same_id=self.same_id)
img_p = load_img(img_p_path)
img_c = load_img(img_c_path)
landmark_p = load_landmark(landmark_p_path)
landmark_c = load_landmark(landmark_c_path)
if self.hflip:
img_p, landmark_p = random_horizonal_flip(img_p, landmark_p)
img_c, landmark_c = random_horizonal_flip(img_c, landmark_c)
if self.resize_crop:
img_p, landmark_p = random_resize_crop(img_p, landmark_p)
img_c, landmark_c = random_resize_crop(img_c, landmark_c)
if self.enlarge:
landmark_c, _ = random_enlarge_landmark(self.landmark_mean, landmark_c)
img_p = self.transform(img_p)
img_c = self.transform(img_c)
field_m2c = cal_field(self.landmark_mean, landmark_c)
field_m2p = cal_field(self.landmark_mean, landmark_p)
field_p2c = cal_field(landmark_p, landmark_c)
item = {
'name': os.path.basename(os.path.dirname(img_p_path)),
'filename': os.path.basename(img_p_path)[:-4],
'img_p': img_p,
'img_c': img_c,
'field_p2c': field_p2c,
'field_m2c': field_m2c,
'field_m2p': field_m2p,
}
else:
img_p_path = self.img_list[index]
img_p = load_img(img_p_path)
name = os.path.basename(os.path.dirname(img_p_path))
img_p = self.transform(img_p)
item = {
'img_p': img_p,
'name': name,
'filename': os.path.basename(img_p_path)[:-4]
}
return item
def __len__(self):
return self.size