tools/inspect_data.py

"""Quick dataloader sanity-check: materialize a handful of batches and print shapes.
Useful to verify audio extraction and csv paths on the real server.

Run:
    python3 tools/inspect_data.py
"""
from __future__ import annotations

import os
import sys
from pathlib import Path

from omegaconf import OmegaConf

sys.path.insert(0, str(Path(__file__).resolve().parent.parent))

from src.data import FairTalkingDataModule


def main():
    cfg_path = Path(__file__).resolve().parent.parent / "configs" / "data" / "fairtalking.yaml"
    data_cfg = OmegaConf.load(cfg_path)
    data_cfg.num_workers = 0
    data_cfg.batch_size = 2
    dm = FairTalkingDataModule(data_cfg=data_cfg)
    dm.setup("fit")

    print(f"train size: {len(dm.train_ds)}")
    print(f"val   size: {len(dm.val_ds)}")

    loader = dm.train_dataloader()
    it = iter(loader)
    for i in range(3):
        b = next(it)
        if b is None:
            print(f"[{i}] empty batch (all samples failed)")
            continue
        print(f"[{i}] video={tuple(b['video'].shape)} audio={tuple(b['audio'].shape)} "
              f"labels={b['label'].tolist()} "
              f"meta_gen={[m['generator'] for m in b['meta']]}")


if __name__ == "__main__":
    main()