CAIR-M3LLM/PMC-VQA-2

Fork of RadGenome/PMC-VQA converted to:

1. Use v2 of the PMC-VQA (train_2.csv and test_2.csv).
2. Wrap images as binary object
3. Remove spacing and option prefix (A:) from questions and choices

Metadata

Name	#train	#val	#test	img#train	img#val	img#test
PMC-VQA	152,603	0	33,430	135,339	0	29,021

Conversion script

from pathlib import Path

from datasets import Dataset, Features, Image, Value

SLAKE_FEAT = {
    "index": Value("uint32"),
    "image": Image(decode=True),
    "Caption": Value("string"),
    "Question": Value("string"),
    "Answer": Value("string"),
} | {f"Choice {op}": Value("string") for op in "ABCD"}


def prefix_strip(s: str, prefix: str) -> str:
    return s.strip().removeprefix(prefix).strip()


def reformat(
    csv_path: Path,
    upload_to: str | None = None,
    image_dir: str = "images",
):
    assert "_2" in csv_path.stem, "Expected a split CSV file with '_2' in the name."
    split = csv_path.stem.replace("_2", "")
    d = Dataset.from_csv(csv_path.as_posix())

    d = d.map(
        lambda e: (
            {
                "image": {"path": f"{image_dir}/{e['Figure_path']}"},
                "Question": e["Question"].strip(),
            }
            | {
                f"Choice {op}": prefix_strip(e[f"Choice {op}"], f"{op}:")
                for op in "ABCD"
            }
        ),
        num_proc=16,
        features=Features(SLAKE_FEAT),
        remove_columns=["Figure_path", "split"],
    )

    if upload_to:
        d.push_to_hub(upload_to, split=split)
    else:
        print(d)
        print(d[0])
        print(f"Would upload to split={split} on the hub.")