File size: 1,758 Bytes
ae419ed | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 | from __future__ import annotations
import argparse
from pathlib import Path
from common import ROOT, load_config
from dynafall.data import load_pickle, save_pickle, split_video_records
from dynafall.features import make_clips, normalize_pose
def main() -> None:
ap = argparse.ArgumentParser()
ap.add_argument("--dataset", required=True)
ap.add_argument("--config", default="configs/default.yaml")
ap.add_argument("--seed", type=int, default=None)
ap.add_argument("--group-key", choices=["video", "scenario"], default="video")
ap.add_argument("--output-name", default=None)
args = ap.parse_args()
cfg = load_config(args.config)
seed = args.seed if args.seed is not None else cfg["seed"]
records = load_pickle(ROOT / "data/poses" / f"{args.dataset}_keypoints.pkl")
splits = split_video_records(records, cfg["splits"], seed, group_key=args.group_key)
out_dir = ROOT / "data/processed" / (args.output_name or args.dataset)
buckets = {k: [] for k in ["train", "val", "test"]}
for rec in records:
split = next(k for k, ids in splits.items() if rec["video_id"] in ids)
norm = normalize_pose(rec["keypoints"])
for i, clip in enumerate(make_clips(norm, cfg["clip_len"], cfg["stride"])):
buckets[split].append({"video_id": rec["video_id"], "clip_id": i, "label": int(rec["label"]), "joint": clip})
for split, rows in buckets.items():
save_pickle(rows, out_dir / f"{split}.pkl")
print(f"{split}: {len(rows)} clips")
save_pickle(
{"seed": seed, "group_key": args.group_key, "splits": {k: sorted(v) for k, v in splits.items()}},
out_dir / "video_splits.pkl",
)
print(f"Wrote {out_dir}")
if __name__ == "__main__":
main()
|