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
|
import os
import numpy as np
import pandas as pd
import random
import PIL
import pytorch_lightning as pl
from PIL import Image
import torch
from torch.utils.data import Dataset, DataLoader, random_split
from torchvision import transforms
class CSVDataModule(pl.LightningDataModule):
def __init__(self,
batch_size,
data_root,
tokenizer,
size=512,
repeats=100,
interpolation="bicubic",
placeholder_token="*",
flip_p=0.5,
center_crop=False):
super().__init__()
self.data_root = data_root
self.tokenizer = tokenizer
self.size = size
self.repeats = repeats
self.placeholder_token = placeholder_token
self.center_crop = center_crop
self.flip_p = flip_p
self.interpolation = interpolation
self.batch_size = batch_size
def prepare_data(self):
metadata = pd.read_csv(f'{self.data_root}/list.csv')
image_paths = [os.path.join(self.data_root, f_path) for f_path in metadata['image'].values]
captions = [caption for caption in metadata['caption'].values]
skips = [skip for skip in metadata['skip'].values]
self.data_full = [(img, cap) for img, cap, skip in zip(image_paths, captions, skips) if skip != "x"]
def setup(self, stage=None):
train_set_size = int(len(self.data_full) * 0.8)
valid_set_size = len(self.data_full) - train_set_size
self.data_train, self.data_val = random_split(self.data_full, [train_set_size, valid_set_size])
train_dataset = CSVDataset(self.data_train, self.tokenizer, size=self.size, repeats=self.repeats, interpolation=self.interpolation,
flip_p=self.flip_p, placeholder_token=self.placeholder_token, center_crop=self.center_crop)
val_dataset = CSVDataset(self.data_val, self.tokenizer, size=self.size, interpolation=self.interpolation,
flip_p=self.flip_p, placeholder_token=self.placeholder_token, center_crop=self.center_crop)
self.train_dataloader_ = DataLoader(train_dataset, batch_size=self.batch_size, shuffle=True)
self.val_dataloader_ = DataLoader(val_dataset, batch_size=self.batch_size)
def train_dataloader(self):
return self.train_dataloader_
def val_dataloader(self):
return self.val_dataloader_
class CSVDataset(Dataset):
def __init__(self,
data,
tokenizer,
size=512,
repeats=1,
interpolation="bicubic",
flip_p=0.5,
placeholder_token="*",
center_crop=False,
):
self.data = data
self.tokenizer = tokenizer
self.num_images = len(self.data)
self._length = self.num_images * repeats
self.placeholder_token = placeholder_token
self.interpolation = {"linear": PIL.Image.LINEAR,
"bilinear": PIL.Image.BILINEAR,
"bicubic": PIL.Image.BICUBIC,
"lanczos": PIL.Image.LANCZOS,
}[interpolation]
self.image_transforms = transforms.Compose(
[
transforms.Resize(size, interpolation=transforms.InterpolationMode.BILINEAR),
transforms.CenterCrop(size) if center_crop else transforms.RandomCrop(size),
transforms.ToTensor(),
transforms.Normalize([0.5], [0.5]),
]
)
self.cache = {}
def __len__(self):
return self._length
def get_example(self, i, flipped):
image_path, text = self.data[i % self.num_images]
if image_path in self.cache:
return self.cache[image_path]
example = {}
image = Image.open(image_path)
if not image.mode == "RGB":
image = image.convert("RGB")
image = self.image_transforms(image)
text = text.format(self.placeholder_token)
example["prompt"] = text
example["input_ids"] = self.tokenizer(
text,
padding="max_length",
truncation=True,
max_length=self.tokenizer.model_max_length,
return_tensors="pt",
).input_ids[0]
example["key"] = "-".join([image_path, "-", str(flipped)])
example["pixel_values"] = image
self.cache[image_path] = example
return example
def __getitem__(self, i):
flipped = random.choice([False, True])
example = self.get_example(i, flipped)
return example
|
