File size: 4,059 Bytes
9ae1b1e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from glob import glob

import torchvision.transforms as transforms
from PIL import Image
from torch.utils.data import Dataset


def make_transform(
    smaller_edge_size: int, patch_size, center_crop=False, max_edge_size=812
) -> transforms.Compose:
    IMAGENET_DEFAULT_MEAN = (0.485, 0.456, 0.406)
    IMAGENET_DEFAULT_STD = (0.229, 0.224, 0.225)
    interpolation_mode = transforms.InterpolationMode.BICUBIC
    assert smaller_edge_size > 0

    if center_crop:
        return transforms.Compose(
            [
                transforms.Resize(
                    size=smaller_edge_size,
                    interpolation=interpolation_mode,
                    antialias=True,
                ),
                transforms.CenterCrop(smaller_edge_size),
                transforms.ToTensor(),
                transforms.Normalize(
                    mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD
                ),
                transforms.Lambda(
                    lambda img: img[
                        :,
                        : min(
                            max_edge_size,
                            (img.shape[1] - img.shape[1] % patch_size),
                        ),
                        : min(
                            max_edge_size,
                            (img.shape[2] - img.shape[2] % patch_size),
                        ),
                    ]
                ),
            ]
        )
    else:
        return transforms.Compose(
            [
                transforms.Resize(
                    size=smaller_edge_size,
                    interpolation=interpolation_mode,
                    antialias=True,
                ),
                transforms.ToTensor(),
                transforms.Normalize(
                    mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD
                ),
                transforms.Lambda(
                    lambda img: img[
                        :,
                        : min(
                            max_edge_size,
                            (img.shape[1] - img.shape[1] % patch_size),
                        ),
                        : min(
                            max_edge_size,
                            (img.shape[2] - img.shape[2] % patch_size),
                        ),
                    ]
                ),
            ]
        )


class VisualDataset(Dataset):
    def __init__(self, transform, imgs=None):
        self.transform = transform
        if imgs is None:
            self.files = [
                'resources/example.jpg',
                'resources/villa.png',
                'resources/000000037740.jpg',
                'resources/000000064359.jpg',
                'resources/000000066635.jpg',
                'resources/000000078420.jpg',
            ]
        else:
            self.files = imgs

    def __len__(self):
        return len(self.files)

    def __getitem__(self, index):
        img = self.files[index]
        img = Image.open(img).convert('RGB')
        if self.transform:
            img = self.transform(img)
        return img


class ImageNetDataset(Dataset):
    def __init__(self, transform, num_train_max=1000000):
        self.transform = transform
        self.files = glob('data/imagenet/train/*/*.JPEG')
        step = len(self.files) // num_train_max
        self.files = self.files[::step]

    def __len__(self):
        return len(self.files)

    def __getitem__(self, index):
        img = Image.open(self.files[index]).convert('RGB')
        img = self.transform(img)
        return img


def load_data(args, model):
    transform = make_transform(
        args.resolution, model.patch_size, center_crop=True
    )
    dataset = ImageNetDataset(
        transform=transform, num_train_max=args.num_train_max
    )
    return dataset


def load_visual_data(args, model):
    transform = make_transform(
        args.visual_size, model.patch_size, max_edge_size=1792
    )
    dataset = VisualDataset(transform=transform, imgs=vars(args).get('imgs'))
    return dataset