-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathdomainnet_dataset.py
132 lines (103 loc) · 7.49 KB
/
domainnet_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
import numpy as np
import torch
from torch.utils.data import Dataset
import torchvision.transforms as transforms
from PIL import Image
import os
class DomainNetDataset(Dataset):
def __init__(self, base_path, site, train=True, transform=None,retrun_index=False):
if train:
self.paths, self.text_labels = np.load('data/DomainNet/{}_train.pkl'.format(site), allow_pickle=True)
else:
self.paths, self.text_labels = np.load('data/DomainNet/{}_test.pkl'.format(site), allow_pickle=True)
label_dict = {'bird':0, 'feather':1, 'headphones':2, 'ice_cream':3, 'teapot':4, 'tiger':5, 'whale':6, 'windmill':7, 'wine_glass':8, 'zebra':9}
self.labels = [label_dict[text] for text in self.text_labels]
self.transform = transform
self.base_path = base_path if base_path is not None else '../data'
self.retrun_index = retrun_index
def __len__(self):
return len(self.labels)
def __getitem__(self, idx):
img_path = os.path.join(self.base_path, self.paths[idx])
label = self.labels[idx]
image = Image.open(img_path)
if len(image.split()) != 3:
image = transforms.Grayscale(num_output_channels=3)(image)
if self.transform is not None:
image = self.transform(image)
if self.retrun_index:
return image, label,idx
else:
return image, label
def prepare_data_domain(args):
net_dataidx_map_train = {}
data_loader_dict = {}
data_base_path = 'data'
transform_train = transforms.Compose([
transforms.Resize([224, 224]),
# transforms.RandomHorizontalFlip(),
# transforms.RandomRotation((-30,30)),
transforms.ToTensor(),
])
transform_test = transforms.Compose([
transforms.Resize([224, 224]),
transforms.ToTensor(),
])
# clipart
clipart_trainset = DomainNetDataset(data_base_path, 'clipart', transform=transform_train,retrun_index=True)
clipart_testset = DomainNetDataset(data_base_path, 'clipart', transform=transform_test, train=False)
# infograph
infograph_trainset = DomainNetDataset(data_base_path, 'infograph', transform=transform_train,retrun_index=True)
infograph_testset = DomainNetDataset(data_base_path, 'infograph', transform=transform_test, train=False)
# painting
painting_trainset = DomainNetDataset(data_base_path, 'painting', transform=transform_train,retrun_index=True)
painting_testset = DomainNetDataset(data_base_path, 'painting', transform=transform_test, train=False)
# quickdraw
quickdraw_trainset = DomainNetDataset(data_base_path, 'quickdraw', transform=transform_train,retrun_index=True)
quickdraw_testset = DomainNetDataset(data_base_path, 'quickdraw', transform=transform_test, train=False)
# real
real_trainset = DomainNetDataset(data_base_path, 'real', transform=transform_train,retrun_index=True)
real_testset = DomainNetDataset(data_base_path, 'real', transform=transform_test, train=False)
# sketch
sketch_trainset = DomainNetDataset(data_base_path, 'sketch', transform=transform_train,retrun_index=True)
sketch_testset = DomainNetDataset(data_base_path, 'sketch', transform=transform_test, train=False)
min_data_len = min(len(clipart_trainset), len(infograph_trainset), len(painting_trainset), len(quickdraw_trainset), len(real_trainset), len(sketch_trainset))
val_len = int(min_data_len * 0.05)
min_data_len = int(min_data_len * 0.05)
clipart_valset = torch.utils.data.Subset(clipart_trainset, list(range(len(clipart_trainset)))[-val_len:])
clipart_trainset = torch.utils.data.Subset(clipart_trainset, list(range(min_data_len)))
infograph_valset = torch.utils.data.Subset(infograph_trainset, list(range(len(infograph_trainset)))[-val_len:])
infograph_trainset = torch.utils.data.Subset(infograph_trainset, list(range(min_data_len)))
painting_valset = torch.utils.data.Subset(painting_trainset, list(range(len(painting_trainset)))[-val_len:])
painting_trainset = torch.utils.data.Subset(painting_trainset, list(range(min_data_len)))
quickdraw_valset = torch.utils.data.Subset(quickdraw_trainset, list(range(len(quickdraw_trainset)))[-val_len:])
quickdraw_trainset = torch.utils.data.Subset(quickdraw_trainset, list(range(min_data_len)))
real_valset = torch.utils.data.Subset(real_trainset, list(range(len(real_trainset)))[-val_len:])
real_trainset = torch.utils.data.Subset(real_trainset, list(range(min_data_len)))
sketch_valset = torch.utils.data.Subset(sketch_trainset, list(range(len(sketch_trainset)))[-val_len:])
sketch_trainset = torch.utils.data.Subset(sketch_trainset, list(range(min_data_len)))
clipart_train_loader = torch.utils.data.DataLoader(clipart_trainset, batch_size=args.batch_size, shuffle=True)
clipart_val_loader = torch.utils.data.DataLoader(clipart_valset, batch_size=args.batch_size, shuffle=False)
clipart_test_loader = torch.utils.data.DataLoader(clipart_testset, batch_size=args.batch_size, shuffle=False)
infograph_train_loader = torch.utils.data.DataLoader(infograph_trainset, batch_size=args.batch_size, shuffle=True)
infograph_val_loader = torch.utils.data.DataLoader(infograph_valset, batch_size=args.batch_size, shuffle=False)
infograph_test_loader = torch.utils.data.DataLoader(infograph_testset, batch_size=args.batch_size, shuffle=False)
painting_train_loader = torch.utils.data.DataLoader(painting_trainset, batch_size=args.batch_size, shuffle=True)
painting_val_loader = torch.utils.data.DataLoader(painting_valset, batch_size=args.batch_size, shuffle=False)
painting_test_loader = torch.utils.data.DataLoader(painting_testset, batch_size=args.batch_size, shuffle=False)
quickdraw_train_loader = torch.utils.data.DataLoader(quickdraw_trainset, batch_size=args.batch_size, shuffle=True)
quickdraw_val_loader = torch.utils.data.DataLoader(quickdraw_valset, batch_size=args.batch_size, shuffle=False)
quickdraw_test_loader = torch.utils.data.DataLoader(quickdraw_testset, batch_size=args.batch_size, shuffle=False)
real_train_loader = torch.utils.data.DataLoader(real_trainset, batch_size=args.batch_size, shuffle=True)
real_val_loader = torch.utils.data.DataLoader(real_valset, batch_size=args.batch_size, shuffle=False)
real_test_loader = torch.utils.data.DataLoader(real_testset, batch_size=args.batch_size, shuffle=False)
sketch_train_loader = torch.utils.data.DataLoader(sketch_trainset, batch_size=args.batch_size, shuffle=True)
sketch_val_loader = torch.utils.data.DataLoader(sketch_valset, batch_size=args.batch_size, shuffle=False)
sketch_test_loader = torch.utils.data.DataLoader(sketch_testset, batch_size=args.batch_size, shuffle=False)
train_loaders = [clipart_train_loader, infograph_train_loader, painting_train_loader, quickdraw_train_loader, real_train_loader, sketch_train_loader]
val_loaders = [clipart_val_loader, infograph_val_loader, painting_val_loader, quickdraw_val_loader, real_val_loader, sketch_val_loader]
test_loaders = [clipart_test_loader, infograph_test_loader, painting_test_loader, quickdraw_test_loader, real_test_loader, sketch_test_loader]
for i in range(len(train_loaders)):
data_loader_dict[i] = {'train_dl_local': train_loaders[i], 'val_dl_local' :val_loaders[i],'test_dl_local' : test_loaders[i]}
net_dataidx_map_train[i] = list(range(len(train_loaders[i].dataset.dataset)))
return data_loader_dict, net_dataidx_map_train