-
Notifications
You must be signed in to change notification settings - Fork 0
/
dataset.py
220 lines (164 loc) · 8.09 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
import torch.utils.data as data
import torchvision.transforms as transforms
import torch
import os
import numpy as np
from PIL import Image
import random
import math
class pgvton_dataset(data.Dataset):
def __init__(self):
super(pgvton_dataset,self).__init__()
def initialize(self,opt):
# dir
self.cloth_mask_dir=os.path.join(opt.data_dir,opt.cloth_mask_dir)
self.human_mask_dir=os.path.join(opt.data_dir,opt.human_mask_dir)
self.densepose_dir = os.path.join(opt.data_dir, opt.densepose_dir)
self.openpose_dir=os.path.join(opt.data_dir,opt.openpose_dir)
# files
self.cloth_mask_files=os.listdir(self.cloth_mask_dir)
self.human_mask_files=os.listdir(self.human_mask_dir)
self.densepose_files=os.listdir(self.densepose_dir)
# num
self.human_sg_num=opt.human_sg_num
self.data_size=len(self.human_mask_files)
self.img_size=opt.img_size
self.joint_num=opt.joint_num
self.miss_value=-1
self.sigma=opt.sigma
# process
transform_list = []
transform_list.append(transforms.Resize(size=self.img_size))
transform_list.append(transforms.ToTensor())
transform_list.append(transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)))
self.trans = transforms.Compose(transform_list)
self.aug=transforms.Compose([transforms.RandomAffine(degrees=8,translate=(0.075, 0.075),scale=(0.9,1.02))])
def __getitem__(self,index):
''' path '''
cloth_mask_path = os.path.join(self.cloth_mask_dir, self.cloth_mask_files[index])
index_random=random.randint(0, self.data_size-1)
cloth_mask_rand_path = os.path.join(self.cloth_mask_dir, self.cloth_mask_files[index_random])
human_mask_path=os.path.join(self.human_mask_dir,self.human_mask_files[index])
densepose_path=os.path.join(self.densepose_dir,self.densepose_files[index])
'''human_mask'''
sg_img = Image.open(human_mask_path)
sg_img = np.expand_dims(np.array(sg_img)[:, :, 0], 0)
sg_img_1d = torch.from_numpy(sg_img).view(-1).long()
ones = torch.sparse.torch.eye(self.human_sg_num) # onehot
ones = ones.index_select(0, sg_img_1d)
sg_onehot = ones.view([self.img_size[0], self.img_size[1], self.human_sg_num])
human_mask= sg_onehot.permute(2, 0, 1)
human_mask_label=torch.from_numpy(sg_img).long()
'''cloth_mask'''
cloth_mask = Image.open(cloth_mask_path)
cloth_mask = np.round(np.array(cloth_mask) / 255)
cloth_mask = np.expand_dims(cloth_mask, 0)
cloth_mask = torch.from_numpy(cloth_mask).long()
cloth_mask_rand= Image.open(cloth_mask_rand_path)
cloth_mask_rand = np.round(np.array(cloth_mask_rand) / 255)
cloth_mask_rand = np.expand_dims(cloth_mask_rand, 0)
cloth_mask_rand = torch.from_numpy(cloth_mask_rand).long()
'''densepose'''
densepose=np.load(densepose_path)
densepose = torch.from_numpy(densepose).long().float()
return human_mask,cloth_mask,cloth_mask_rand,densepose,human_mask_label
def __len__(self):
return self.data_size
def joint2map(self,J):
joint=np.zeros(shape=(2,self.joint_num))
joint[0,:]=np.array(J[0][:])
joint[1, :] = np.array(J[1][:])
joint=joint.astype(float)
map=np.zeros(shape=(self.img_size[0],self.img_size[1],self.joint_num),dtype='float32')
for i in range(joint.shape[1]):
if joint[0,i]==self.miss_value or joint[1,i]==self.miss_value:
continue
joint_x=int(joint[0,i])
joint_y=int(joint[1,i])
xx,yy = np.meshgrid(np.arange(self.img_size[1]), np.arange(self.img_size[0]))
map[:,:,i]=np.exp(-((yy - joint_y) ** 2 + (xx - joint_x) ** 2) / (2 * self.sigma ** 2))
return map
class RandomErasing(object):
""" Randomly selects a rectangle region in an image and erases its pixels.
Args:
probability: The probability that the Random Erasing operation will be performed.
sl: Minimum proportion of erased area against input image.擦除面积与输入图像的最小比例
sh: Maximum proportion of erased area against input image.
r1: Minimum aspect ratio of erased area.擦除面积的最小纵横比
mean: Erasing value.
"""
def __init__(self, probability=0.1, sl=0.02, sh=0.50, r1=0.3, mean=0.0):
self.probability = probability
self.mean = mean
self.sl = sl
self.sh = sh
self.r1 = r1
def __call__(self, img):
for attempt in range(20):
if random.uniform(0, 1) >= self.probability :
area = img.size()[1] * img.size()[2]
target_area = random.uniform(self.sl, self.sh) * area
aspect_ratio = random.uniform(self.r1, 1 / self.r1)
h = int(round(math.sqrt(target_area * aspect_ratio)))
w = int(round(math.sqrt(target_area / aspect_ratio)))
if w < img.size()[2] and h < img.size()[1]:
x1 = random.randint(0, img.size()[1] - h)
y1 = random.randint(0, img.size()[2] - w)
if img.size()[0] == 7:
img[3, x1:x1 + h, y1:y1 + w] = self.mean
else:
img[3, x1:x1 + h, y1:y1 + w] = self.mean
return img
class armpaint_dataset(data.Dataset):
def __init__(self):
super(armpaint_dataset,self).__init__()
def initialize(self,opt):
# dir
self.human_img_dir=os.path.join(opt.data_dir,opt.human_dir)
self.human_mask_dir=os.path.join(opt.data_dir,opt.human_mask_dir)
self.human_mask_src_dir = os.path.join(opt.data_dir, opt.human_mask_src_dir)
self.mode=opt.mode
# files
self.human_img_files=os.listdir(self.human_img_dir)
self.human_mask_files=os.listdir(self.human_mask_dir)
self.human_mask_src_files = os.listdir(self.human_mask_src_dir)
# num
self.human_sg_num=opt.human_sg_num
self.data_size=len(self.human_img_files)
self.img_size=opt.img_size
# process
transform_list = []
transform_list.append(transforms.Resize(size=self.img_size))
transform_list.append(transforms.ToTensor())
transform_list.append(transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)))
self.trans = transforms.Compose(transform_list)
self.aug = transforms.Compose([RandomErasing()])
def __getitem__(self,index):
''' path '''
human_img_path=os.path.join(self.human_img_dir,self.human_img_files[index])
human_mask_path=os.path.join(self.human_mask_dir,self.human_mask_files[index])
human_mask_src_path = os.path.join(self.human_mask_src_dir, self.human_mask_src_files[index])
human_img=Image.open(human_img_path)
human_img=self.trans(human_img)
sg_img = Image.open(human_mask_path)
sg_img = np.expand_dims(np.array(sg_img)[:, :, 0], 0)
sg_img_1d = torch.from_numpy(sg_img).view(-1).long()
ones = torch.sparse.torch.eye(self.human_sg_num) # onehot
ones = ones.index_select(0, sg_img_1d)
sg_onehot = ones.view([self.img_size[0], self.img_size[1], self.human_sg_num])
human_mask= sg_onehot.permute(2, 0, 1)
sg_img_src = Image.open(human_mask_src_path)
sg_img_src = np.expand_dims(np.array(sg_img_src)[:, :, 0], 0)
sg_img_1d_src = torch.from_numpy(sg_img_src).view(-1).long()
ones_src = torch.sparse.torch.eye(self.human_sg_num) # onehot
ones_src = ones_src.index_select(0, sg_img_1d_src)
sg_onehot_src = ones_src.view([self.img_size[0], self.img_size[1], self.human_sg_num])
human_mask_src = sg_onehot_src.permute(2, 0, 1)
if self.mode=="train":
human_mask_eras=self.aug(human_mask.clone())
else:
human_mask_eras = human_mask_src
human_mask_diff=human_mask.clone()-human_mask_eras.clone()
return human_img,human_mask,human_mask_eras,human_mask_diff
def __len__(self):
return self.data_size