Datasets:

wanderkid

/

UniMER_Dataset

like 9

@AdithyaSK
In our dataset loading process, there are some LaTeX annotations that cannot be rendered into images. As a result, the number of items in the original annotation file train.txt and the number of images in the images folder do not exactly match. To align the data, you can match the image filenames directly with the line numbers in the text files. Please note that the image named 0000000.png corresponds to the first line of the text file. You can refer to our dataset code for specifics.

class Im2LatexDataset(BaseDataset):

    def init_samples(self):
        samples = []
        for vis_root, anno_path in zip(self.vis_root, self.anno_path):
            images = [path.replace('\\', '/') for path in glob.glob(osp.join(vis_root, '*.png'))]
            indices = [int(osp.basename(img).split('.')[0]) for img in images]

            eqs = open(anno_path, 'r').read().split('\n')
            eqs = [eqs[_] for _ in indices]

            for i, e in zip(images, eqs):
                samples.append({"image": i, "equation": e, "vis_root": vis_root})
        return samples

    def __getitem__(self, index):
        ann = self.samples[index]
        try:
            image = self.vis_processor(self._read_image(ann))
        except Exception:
            return self[(index + 1) % len(self)]
        if image is None:
            return self[(index + 1) % len(self)]
        equation = ann["equation"]
        return {"image": image, "text_input": equation, "id": index}

    def _read_image(self, sample, image_key="image"):
        img_file = sample[image_key]
        vis_root = sample["vis_root"]
        image_path = osp.join(vis_root, img_file)
        image = self.reader['body'](image_path)
        if isinstance(image, bytes):
            bytes_stream = BytesIO(image)
            image = Image.open(bytes_stream)
        image = image.convert("RGB")
        return image

    def init_reader(self):
        if not isinstance(self.vis_root, str):
            vis_root = self.vis_root[0]
        else:
            vis_root = self.vis_root
        if vis_root.startswith('cluster'):
            from petrel_client.client import Client
            client = Client("~/petreloss.conf")
            reader = {'type': 'PetrelReader', 'body': client.get}
        else:
            reader = {'type': 'LocalReader', 'body': Image.open}
        return reader

    def collater(self, samples):
        image_list, question_list, id_list = [], [], []

        for sample in samples:
            image_list.append(sample["image"])
            question_list.append(sample["text_input"])
            id_list.append(sample["id"])

        return {
            "image": torch.stack(image_list, dim=0),
            "text_input": question_list,
            "id": id_list
        }

Is it possible to get the dataset you used for training other models such as MFD, LayoutLM etc as well? I left a comment under the issues section for the same. I want to play around with the model finetuning/training myself as well. Thanks.

Unfortunately, the dataset used for training models such as MFD and LayoutLM is in-house and not authorized for public release at this time.