Spaces:
Build error
Build error
File size: 8,436 Bytes
a4fb052 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 |
import pandas as pd
import os
import torch
from PIL import Image
from torch.utils.data import Dataset
import clip
from torch.utils.data import DataLoader
import torchvision.transforms as tf
import torchvision.transforms.functional as TF
try:
from torchvision.transforms import InterpolationMode
BICUBIC = InterpolationMode.BICUBIC
except ImportError:
BICUBIC = Image.BICUBIC
class ExtractFeaturesDataset(Dataset):
def __init__(self,
annotations,
img_path,
image_transforms=None,
question_transforms=None,
tta=False):
self.img_path = img_path
self.image_transforms = image_transforms
self.question_transforms = question_transforms
self.img_ids = annotations["image_id"].values
self.split = annotations["split"].values
self.questions = annotations["question"].values
self.tta = tta
def __getitem__(self, index):
image_id = self.img_ids[index]
split = self.split[index]
# image input
with open(os.path.join(self.img_path, split, image_id), "rb") as f:
img = Image.open(f)
if self.tta:
image_augmentations = []
for transform in self.image_transforms:
image_augmentations.append(transform(img))
img = torch.stack(image_augmentations, dim=0)
else:
img = self.image_transforms(img)
question = self.questions[index]
if self.question_transforms:
question = self.question_transforms(question)
# question input
question = clip.tokenize(question, truncate=True)
question = question.squeeze()
return img, question, image_id
def __len__(self):
return len(self.img_ids)
def _convert_image_to_rgb(image):
return image.convert("RGB")
def Sharpen(sharpness_factor=1.0):
def wrapper(x):
return TF.adjust_sharpness(x, sharpness_factor)
return wrapper
def Rotate(angle=0.0):
def wrapper(x):
return TF.rotate(x, angle)
return wrapper
def transform_crop(n_px):
return tf.Compose([
tf.Resize(n_px, interpolation=BICUBIC),
tf.CenterCrop(n_px),
_convert_image_to_rgb,
tf.ToTensor(),
tf.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711)),
])
def transform_crop_rotate(n_px, rotation_angle=0.0):
return tf.Compose([
Rotate(angle=rotation_angle),
tf.Resize(n_px, interpolation=BICUBIC),
tf.CenterCrop(n_px),
_convert_image_to_rgb,
tf.ToTensor(),
tf.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711)),
])
def transform_resize(n_px):
return tf.Compose([
tf.Resize((n_px, n_px), interpolation=BICUBIC),
_convert_image_to_rgb,
tf.ToTensor(),
tf.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711)),
])
def transform_resize_rotate(n_px, rotation_angle=0.0):
return tf.Compose([
Rotate(angle=rotation_angle),
tf.Resize((n_px, n_px), interpolation=BICUBIC),
_convert_image_to_rgb,
tf.ToTensor(),
tf.Normalize((0.48145466, 0.4578275, 0.40821073), (0.26862954, 0.26130258, 0.27577711)),
])
def get_tta_preprocess(img_size):
img_preprocess = [
transform_crop(img_size),
transform_crop_rotate(img_size, rotation_angle=90.0),
transform_crop_rotate(img_size, rotation_angle=270.0),
transform_resize(img_size),
transform_resize_rotate(img_size, rotation_angle=90.0),
transform_resize_rotate(img_size, rotation_angle=270.0),
]
return img_preprocess
def question_preprocess(question, debug=False):
question = question.replace("?", ".")
if question[-1] == " ":
question = question[:-1]
if question[-1] != ".":
question = question + "."
if debug:
print("Question:", question)
return question
def get_dataloader_extraction(config):
if config.use_question_preprocess:
print("Using custom preprocessing: Question")
question_transforms = question_preprocess
else:
question_transforms = None
if config.tta:
("Using augmentation transforms:")
img_preprocess = get_tta_preprocess(config.img_size)
else:
("Using original CLIP transforms:")
img_preprocess = transform_crop(config.img_size)
train_data = pd.read_csv(config.train_annotations_path)
train_dataset = ExtractFeaturesDataset(annotations = train_data,
img_path=config.img_path,
image_transforms=img_preprocess,
question_transforms=question_transforms,
tta=config.tta)
train_loader = DataLoader(dataset=train_dataset,
batch_size=config.batch_size,
shuffle=False,
num_workers=config.num_workers)
test_data = pd.read_csv(config.test_annotations_path)
test_dataset = ExtractFeaturesDataset(annotations = test_data,
img_path=config.img_path,
image_transforms=img_preprocess,
question_transforms=question_transforms,
tta=config.tta)
test_loader = ExtractFeaturesDataset(dataset=test_dataset,
batch_size=config.batch_size,
shuffle=False,
num_workers=config.num_workers)
return train_loader, test_loader
def get_dataloader_inference(config):
if config.use_question_preprocess:
print("Using custom preprocessing: Question")
question_transforms = question_preprocess
else:
question_transforms = None
if config.tta:
("Using augmentation transforms:")
img_preprocess = transform_resize(config.img_size)
else:
("Using original CLIP transforms:")
img_preprocess = transform_crop(config.img_size)
train_data = pd.read_csv(config.train_annotations_path)
train_dataset = ExtractFeaturesDataset(annotations = train_data,
img_path=config.img_path,
image_transforms=img_preprocess,
question_transforms=question_transforms,
tta=config.tta)
train_loader = DataLoader(dataset=train_dataset,
batch_size=config.batch_size,
shuffle=False,
num_workers=config.num_workers)
test_data = pd.read_csv(config.test_annotations_path)
test_dataset = ExtractFeaturesDataset(annotations = test_data,
img_path=config.img_path,
image_transforms=img_preprocess,
question_transforms=question_transforms,
tta=config.tta)
test_loader = ExtractFeaturesDataset(dataset=test_dataset,
batch_size=config.batch_size,
shuffle=False,
num_workers=config.num_workers)
return train_loader, test_loader
|