scripts/download_dataset.py

from datasets import load_dataset
import argparse
from pathlib import Path
import json


def process_batch(examples):
    return examples


if __name__ == "__main__":
    # Create argument parser
    parser = argparse.ArgumentParser(description="Download dataset.")
    parser.add_argument("--base_dir", required=True, help="Directory to save the downloaded files.")
    args = parser.parse_args()

    base_dir = Path(args.base_dir)
    base_dir.mkdir(exist_ok=True, parents=True)

    # Load dataset
    ds = load_dataset(
        "gvecchio/MatSynth",
        streaming=True,
    )

    # Setup dummy processing
    ds = ds.map(process_batch, batched=False, batch_size=1)

    for split in ds:
        for item in ds[split]:
            name = item["name"]
            dest_dir = base_dir / split / item["metadata"]["category"] / name
            dest_dir.mkdir(exist_ok=True, parents=True)

            # Save metadata
            with open(dest_dir / "metadata.json", "w") as f:
                item["metadata"]["physical_size"] = str(
                    item["metadata"]["physical_size"]
                )
                json.dump(item["metadata"], f, indent=4)

            # Save images
            item["basecolor"].save(dest_dir / "basecolor.png")
            item["diffuse"].save(dest_dir / "diffuse.png")
            item["displacement"].save(dest_dir / "displacement.png")
            item["specular"].save(dest_dir / "specular.png")
            item["height"].save(dest_dir / "height.png")
            item["metallic"].save(dest_dir / "metallic.png")
            item["normal"].save(dest_dir / "normal.png")
            item["opacity"].save(dest_dir / "opacity.png")
            item["roughness"].save(dest_dir / "roughness.png")
            if item["blend_mask"] is not None:
                item["blend_mask"].save(dest_dir / "blend_mask.png")