unitable/src/datamodule/fintabnet.py

from typing import Any, Literal, Union
from pathlib import Path
import jsonlines
from PIL import Image
from torch import Tensor
from torch.utils.data import Dataset
import torchvision.transforms as transforms


class FinTabNet(Dataset):
    """Load PubTabNet for different training purposes."""

    def __init__(
        self,
        root_dir: Union[Path, str],
        label_type: Literal["image", "html", "cell", "bbox"],
        transform: transforms = None,
        jsonl_filename: Union[Path, str] = None,
    ) -> None:
        super().__init__()

        self.root_dir = Path(root_dir)
        self.label_type = label_type
        self.transform = transform

        if label_type != "image":
            jsonl_file = self.root_dir / jsonl_filename
            with jsonlines.open(jsonl_file) as f:
                self.image_label_pair = list(f)

    def __len__(self):
        return len(self.image_label_pair)

    def __getitem__(self, index: int) -> Any:
        if self.label_type == "image":
            raise ValueError("FinTabNet is not used in pretraining.")
        else:
            obj = self.image_label_pair[index]
            img_name = f"{obj['table_id']}.png"
            img = Image.open(self.root_dir / "image" / img_name)
            if self.transform:
                img = self.transform(img)

            sample = dict(filename=obj["filename"], image=img)

            if self.label_type == "html":
                sample["html"] = obj["html"]["structure"]["tokens"]
                return sample
            else:
                raise ValueError("Task not supported in current dataset.")