File size: 2,722 Bytes
6eb58ae |
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 |
import os
import json
import random
from torch.utils.data import Dataset
from torchvision.datasets.utils import download_url
from PIL import Image
from data.utils import pre_caption
class nlvr_dataset(Dataset):
def __init__(self, transform, image_root, ann_root, split):
'''
image_root (string): Root directory of images
ann_root (string): directory to store the annotation file
split (string): train, val or test
'''
urls = {'train':'https://storage.googleapis.com/sfr-vision-language-research/datasets/nlvr_train.json',
'val':'https://storage.googleapis.com/sfr-vision-language-research/datasets/nlvr_dev.json',
'test':'https://storage.googleapis.com/sfr-vision-language-research/datasets/nlvr_test.json'}
filenames = {'train':'nlvr_train.json','val':'nlvr_dev.json','test':'nlvr_test.json'}
download_url(urls[split],ann_root)
self.annotation = json.load(open(os.path.join(ann_root,filenames[split]),'r'))
self.transform = transform
self.image_root = image_root
def __len__(self):
return len(self.annotation)
def __getitem__(self, index):
ann = self.annotation[index]
image0_path = os.path.join(self.image_root,ann['images'][0])
image0 = Image.open(image0_path).convert('RGB')
image0 = self.transform(image0)
image1_path = os.path.join(self.image_root,ann['images'][1])
image1 = Image.open(image1_path).convert('RGB')
image1 = self.transform(image1)
sentence = pre_caption(ann['sentence'], 40)
if ann['label']=='True':
label = 1
else:
label = 0
words = sentence.split(' ')
if 'left' not in words and 'right' not in words:
if random.random()<0.5:
return image0, image1, sentence, label
else:
return image1, image0, sentence, label
else:
if random.random()<0.5:
return image0, image1, sentence, label
else:
new_words = []
for word in words:
if word=='left':
new_words.append('right')
elif word=='right':
new_words.append('left')
else:
new_words.append(word)
sentence = ' '.join(new_words)
return image1, image0, sentence, label
|