preprocessing/make_manifest.py
| 1 | """Write the encoder manifest (episodes.jsonl) for one task from the downloaded raw/ tree. |
| 2 | |
| 3 | python preprocessing/make_manifest.py --raw-root $ACTS_DATA_ROOT/raw --task rope --out OUT/episodes_rope.jsonl |
| 4 | |
| 5 | raw/ layout (WitneyWW/vtm-cosmos-data): |
| 6 | raw/videos/<episode>/{view_middle,tactile_left,tactile_right,wrist_left,wrist_right}.mp4 (30 fps) |
| 7 | raw/parquet/<episode>.parquet per-native-frame sensor poses + calibrated normal force |
| 8 | raw/episodes.json [{episode, tree, num_frames}] for the 166 split episodes |
| 9 | Row schema = what third_party/wan22_vae/encode_tactile_wan22.py reads.""" |
| 10 | from __future__ import annotations |
| 11 | |
| 12 | import argparse |
| 13 | import json |
| 14 | from pathlib import Path |
| 15 | |
| 16 | TREES = {"umi_all": "mixed_mb", "rope": "mixed_rope", "pusht": "mixed_pushT"} |
| 17 | STREAMS = ["view_middle", "tactile_left", "tactile_right", "wrist_left", "wrist_right"] |
| 18 | |
| 19 | |
| 20 | def main(): |
| 21 | ap = argparse.ArgumentParser() |
| 22 | ap.add_argument("--raw-root", type=Path, required=True) |
| 23 | ap.add_argument("--task", required=True, choices=sorted(TREES)) |
| 24 | ap.add_argument("--out", type=Path, required=True) |
| 25 | ap.add_argument("--episodes", nargs="*", default=None, help="subset (episode stems)") |
| 26 | a = ap.parse_args() |
| 27 | rows = [r for r in json.loads((a.raw_root / "episodes.json").read_text()) if r["tree"] == TREES[a.task]] |
| 28 | if a.episodes: |
| 29 | rows = [r for r in rows if r["episode"] in set(a.episodes)] |
| 30 | out = [] |
| 31 | for r in rows: |
| 32 | d = a.raw_root / "videos" / r["episode"] |
| 33 | row = {"episode": r["episode"], "num_frames": int(r["num_frames"]), "task": a.task} |
| 34 | for s in STREAMS: |
| 35 | p = d / f"{s}.mp4" |
| 36 | if not p.is_file(): |
| 37 | raise FileNotFoundError(p) |
| 38 | row[s] = str(p.resolve()) |
| 39 | out.append(row) |
| 40 | a.out.parent.mkdir(parents=True, exist_ok=True) |
| 41 | a.out.write_text("".join(json.dumps(r) + "\n" for r in out)) |
| 42 | print(f"{len(out)} episodes -> {a.out}") |
| 43 | |
| 44 | |
| 45 | if __name__ == "__main__": |
| 46 | main() |
| 47 | |