def __getitem__()

in src/data_loader.py [0:0]

• 40 lines of code
• 8 McCabe index (conditional complexity)

    def __getitem__(self, index):
        """Returns one data pair (image and caption)."""

        sample = self.dataset[self.ids[index]]
        img_id = sample['id']
        paths = sample['images'][0:self.maxnumims]

        idx = index

        labels = self.dataset[self.ids[idx]]['ingredients']

        true_ingr_idxs = []
        for i in range(len(labels)):
            true_ingr_idxs.append(self.ingrs_vocab(labels[i]))
        true_ingr_idxs = list(set(true_ingr_idxs))

        if self.shuffle:
            np.random.shuffle(true_ingr_idxs)

        if self.include_eos:
            true_ingr_idxs.append(self.ingrs_vocab('<end>'))

        if len(paths) == 0:
            path = None
            image_input = torch.zeros((3, 224, 224))
        else:
            if self.split == 'train':
                img_idx = np.random.randint(0, len(paths))
            else:
                img_idx = 0
            path = paths[img_idx]
            impath = os.path.join(self.root, path[0], path[1], path[2], path[3], path)
            if self.use_lmdb:
                try:
                    with self.image_file.begin(write=False) as txn:
                        image = txn.get(path.encode())
                        image = np.fromstring(image, dtype=np.uint8)
                        image = np.reshape(image, (256, 256, 3))
                    image = Image.fromarray(image.astype('uint8'), 'RGB')
                except:
                    print("Image id not found in lmdb. Loading jpeg file...")
                    image = Image.open(impath).convert('RGB')
            else:
                image = Image.open(impath).convert('RGB')
            if self.transform is not None:
                image = self.transform(image)
            image_input = image

        return image_input, true_ingr_idxs