def resnet50(pretrained=False, **kwargs): model = ResNet(MyBottleneck, [3, 4, 6, 3], **kwargs) if pretrained: model.load_state_dict(load(pretrained)) model.eval() return model
short_name, extension = os.path.splitext(img_file_name) img_id_int = int(short_name.split('_')[-1]) img_id = str(img_id_int) img = os.path.join(self.root, 'vqa', 'image', self.split, img_file_name) img = Image.open(img).convert('RGB') return img_id, transform(img) def __len__(self): return self.length batch_size = 50 resnet = ResNet().to(device) resnet.eval() def create_dataset(split): dataset = VQAImg(vqa_root, split) dataloader = DataLoader(dataset, batch_size=batch_size, num_workers=4) total_img_num = len(dataset) print('found {} images in {} split ...'.format(total_img_num, split)) f = h5py.File('data/vqa/spatial/{}_spatial.hdf5'.format(split), 'w', libver='latest') dset = f.create_dataset('data', (total_img_num, 2048, 7, 7), dtype='f4') info = {}