example_usage.py

"""Minimal usage example for DroidRGBDataset.

Run from the `droid_share/` directory:
    python example_usage.py --root /path/to/lerobot/droid_1.0.1
"""

import argparse
from pathlib import Path

import torch
from torch.utils.data import DataLoader

from droid_rgb_dataset import DroidRGBDataset


def main() -> None:
    parser = argparse.ArgumentParser()
    parser.add_argument("--root", required=True, help="Path to lerobot/droid_1.0.1")
    parser.add_argument("--stats-dir", default=str(Path(__file__).parent / "_stats"))
    parser.add_argument("--n-frames", type=int, default=8)
    parser.add_argument("--stride", type=int, default=3, help="3 = 5Hz, 1 = 15Hz")
    parser.add_argument("--batch-size", type=int, default=2)
    parser.add_argument("--num-workers", type=int, default=2)
    parser.add_argument("--save-preview", default="preview.png")
    args = parser.parse_args()

    dataset = DroidRGBDataset(
        root=args.root,
        stats_dir=args.stats_dir,
        camera_keys=(
            "observation.images.exterior_2_left",
            "observation.images.wrist_left",
        ),
        n_frames=args.n_frames,
        stride=args.stride,
        image_size=(224, 224),
    )
    print(f"usable episodes: {len(dataset)}")

    loader = DataLoader(
        dataset,
        batch_size=args.batch_size,
        num_workers=args.num_workers,
        shuffle=True,
        collate_fn=_collate,
    )

    batch = next(iter(loader))
    ext = batch["observation.images.exterior_2_left"]
    wrist = batch["observation.images.wrist_left"]
    print(f"exterior: shape={tuple(ext.shape)} dtype={ext.dtype}")
    print(f"wrist:    shape={tuple(wrist.shape)} dtype={wrist.dtype}")
    print(f"episode_index: {batch['episode_index'].tolist()}")
    print(f"start_frame:   {batch['start_frame'].tolist()}")

    try:
        from PIL import Image
        import numpy as np
    except ImportError:
        return
    grid = torch.cat([ext[0], wrist[0]], dim=2).numpy()  # (T, H, 2W, 3)
    row = np.concatenate(list(grid), axis=1)  # (H, T*2W, 3)
    Image.fromarray(row).save(args.save_preview)
    print(f"saved preview: {args.save_preview}")


def _collate(samples):
    out = {}
    for k in samples[0]:
        if isinstance(samples[0][k], torch.Tensor):
            out[k] = torch.stack([s[k] for s in samples], dim=0)
        else:
            out[k] = torch.tensor([s[k] for s in samples])
    return out


if __name__ == "__main__":
    main()