from __future__ import annotations import argparse import json import subprocess from pathlib import Path import numpy as np import pandas as pd from tqdm import tqdm from psi.data.seed.motion_repr import build_seed_textop_action def load_motion_array(motion_dir: Path) -> np.ndarray: joint = pd.read_csv(motion_dir / "joint_pos.csv").to_numpy(dtype=np.float32) body_pos = pd.read_csv(motion_dir / "body_pos.csv").to_numpy(dtype=np.float32) body_quat = pd.read_csv(motion_dir / "body_quat.csv").to_numpy(dtype=np.float32) length = min(joint.shape[0], body_pos.shape[0], body_quat.shape[0]) return np.concatenate([joint[:length], body_pos[:length], body_quat[:length]], axis=1).astype(np.float32) def count_csv_rows(path: Path) -> int: with path.open("r", encoding="utf-8", errors="ignore") as f: return max(0, sum(1 for _ in f) - 1) def has_audio_stream(video_path: Path) -> bool: # 统计文件与训练集保持一致:没有音频流的视频对应motion不参与统计。 cmd = [ "ffprobe", "-v", "error", "-select_streams", "a:0", "-show_entries", "stream=index", "-of", "csv=p=0", str(video_path), ] proc = subprocess.run(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=False) return proc.returncode == 0 and bool(proc.stdout.strip()) def stats_from_arrays(array: np.ndarray) -> dict[str, list[float]]: return { "mean": np.mean(array, axis=0).tolist(), "std": np.std(array, axis=0).tolist(), "min": np.min(array, axis=0).tolist(), "max": np.max(array, axis=0).tolist(), "q01": np.quantile(array, 0.01, axis=0).tolist(), "q99": np.quantile(array, 0.99, axis=0).tolist(), } def main() -> None: parser = argparse.ArgumentParser(description="Calculate 70D TextOp stats for dance audio->motion data.") parser.add_argument("--motion-root", type=str, required=True) parser.add_argument("--video-root", type=str, default="") parser.add_argument("--output-path", type=str, required=True) parser.add_argument("--action-chunk-size", type=int, default=30) parser.add_argument("--num-past-frames", type=int, default=0) args = parser.parse_args() motion_root = Path(args.motion_root) video_root = Path(args.video_root) if args.video_root else None motion_dirs = sorted([p for p in motion_root.iterdir() if p.is_dir()]) if not motion_dirs: raise RuntimeError(f"No motion dirs found in {motion_root}") actions = [] states = [] valid_motion_dirs = [] skipped_no_audio = 0 for motion_dir in tqdm(motion_dirs, desc="Collecting dance TextOp stats"): # 如果提供video-root,则只统计确实有音频输入的motion。 if video_root is not None: video_path = video_root / f"{motion_dir.name}.mp4" if not video_path.exists(): raise FileNotFoundError(f"Matched video file does not exist: {video_path}") if not has_audio_stream(video_path): skipped_no_audio += 1 continue joint_path = motion_dir / "joint_pos.csv" if not joint_path.exists() or count_csv_rows(joint_path) <= 1: continue motion = load_motion_array(motion_dir) if motion.shape[0] <= 1: continue actions.append(build_seed_textop_action(motion[1:], motion[:1])) states.append(motion) valid_motion_dirs.append(motion_dir) if not actions: raise RuntimeError(f"No valid motion rows found in {motion_root}") action_array = np.concatenate(actions, axis=0).astype(np.float32) state_array = np.concatenate(states, axis=0).astype(np.float32) stats = { "action": stats_from_arrays(action_array), "states": stats_from_arrays(state_array), "meta": { "dataset": "psi0_dance_gvhmr_audio", "representation": "joint_abs29_joint_delta29_root_delta_local3_root_rot6d_delta6_root_height1_foot_contact2", "stat_mode": "frame_level_motion_root_audio_filtered" if video_root is not None else "frame_level_motion_root", "motion_root": str(motion_root), "video_root": str(video_root) if video_root is not None else "", "skipped_no_audio": int(skipped_no_audio), "action_dim": 70, "state_dim": 36, "action_chunk_size": int(args.action_chunk_size), "num_past_frames": int(args.num_past_frames), "num_motions": int(len(valid_motion_dirs)), "num_action_rows": int(action_array.shape[0]), "num_state_rows": int(state_array.shape[0]), }, } output_path = Path(args.output_path) output_path.parent.mkdir(parents=True, exist_ok=True) output_path.write_text(json.dumps(stats, indent=2), encoding="utf-8") print(f"Wrote {output_path}") print(json.dumps(stats["meta"], indent=2)) if __name__ == "__main__": main()