batch_visualize.py

#!/usr/bin/env python3
"""
批量可视化 contact / force 并拼成 combined 三联视频。

对 --root 下每个含 modalities/ 的 episode:
  1) 渲染 contact_vis.mp4 / force_vis.mp4 (黑底, 对齐原视频)
  2) 横向拼接 original + contact + force -> combined_vis.mp4

一遍 pass 同时写出三个 vis 文件 + combined, 不重复解码。
多进程并行, 已有 combined 默认跳过, 单 episode 失败不中断整批。

用法:
  python batch_visualize.py --root .../all_episodes
  python batch_visualize.py --root .../all_episodes --workers 8 --overwrite
  python batch_visualize.py --root .../all_episodes --glob "00*"   # 只跑匹配的
"""
import os
import json
import argparse
import traceback
import numpy as np
import cv2
from pathlib import Path
from concurrent.futures import ProcessPoolExecutor, as_completed

LABELS = ["original", "contact", "force"]


def load_norm(modalities_dir):
    with open(modalities_dir / "norm_params.json") as f:
        n = json.load(f)
    return n["max_contact"], n["max_force"]


def render_contact(contact, max_contact):
    """黑底白斑: 亮度 = 归一化接触强度"""
    c = np.clip(contact / max_contact, 0, 1)
    gray = (c * 255).astype(np.uint8)
    return cv2.cvtColor(gray, cv2.COLOR_GRAY2BGR)


def render_force(force, max_force):
    """黑底彩斑: 亮度 = |force|, 色相 = xyz 方向"""
    mag = np.sqrt((force ** 2).sum(-1))
    brightness = np.clip(mag / max_force, 0, 1)
    norm = mag.copy()
    norm[norm < 1e-8] = 1.0
    direction = force / norm[..., None]          # 单位向量 [-1,1]
    color = (direction + 1.0) / 2.0              # -> [0,1] 当 RGB
    rgb = color * brightness[..., None]
    bgr = (rgb[..., ::-1] * 255).astype(np.uint8)
    return bgr


def _fit(img, W, H):
    """模态分辨率万一和原视频不一致, resize 回去避免 hstack 崩"""
    if img.shape[1] != W or img.shape[0] != H:
        img = cv2.resize(img, (W, H), interpolation=cv2.INTER_NEAREST)
    return img


def process_episode(ep, video_name, overwrite):
    ep = Path(ep)
    mod = ep / "modalities"
    combined_path = ep / "combined_vis.mp4"

    if not mod.is_dir():
        return ep.name, "skip (no modalities/)"
    if combined_path.exists() and not overwrite:
        return ep.name, "skip (exists)"

    max_contact, max_force = load_norm(mod)

    cap = cv2.VideoCapture(str(ep / video_name))
    if not cap.isOpened():
        cap.release()
        return ep.name, f"skip (cannot open {video_name})"
    fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
    W = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
    H = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))

    fourcc = cv2.VideoWriter_fourcc(*"mp4v")
    c_out = cv2.VideoWriter(str(mod / "contact" / "contact_vis.mp4"), fourcc, fps, (W, H))
    f_out = cv2.VideoWriter(str(mod / "force" / "force_vis.mp4"), fourcc, fps, (W, H))
    comb = cv2.VideoWriter(str(combined_path), fourcc, fps, (W * 3, H))

    frame_idx = 0
    while True:
        ret, orig = cap.read()
        if not ret:
            break

        cp = mod / "contact" / f"{frame_idx:06d}.npy"
        c_vis = render_contact(np.load(cp), max_contact) if cp.exists() \
            else np.zeros((H, W, 3), np.uint8)
        c_vis = _fit(c_vis, W, H)

        fp = mod / "force" / f"{frame_idx:06d}.npy"
        f_vis = render_force(np.load(fp), max_force) if fp.exists() \
            else np.zeros((H, W, 3), np.uint8)
        f_vis = _fit(f_vis, W, H)

        c_out.write(c_vis)
        f_out.write(f_vis)

        trio = [orig.copy(), c_vis.copy(), f_vis.copy()]
        for img, label in zip(trio, LABELS):
            cv2.putText(img, label, (10, 25),
                        cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 0), 2)
        comb.write(np.hstack(trio))
        frame_idx += 1

    cap.release()
    c_out.release()
    f_out.release()
    comb.release()
    return ep.name, f"ok ({frame_idx} frames)"


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--root", type=str, required=True,
                        help="包含多个 episode 子目录的父目录")
    parser.add_argument("--glob", type=str, default="*",
                        help="episode 目录名匹配模式 (相对 root)")
    parser.add_argument("--video_name", type=str, default="camera2.mp4")
    parser.add_argument("--workers", type=int,
                        default=min(8, os.cpu_count() or 1))
    parser.add_argument("--overwrite", action="store_true",
                        help="重新生成已存在的 combined_vis.mp4")
    args = parser.parse_args()

    root = Path(args.root)
    ep_dirs = [p for p in sorted(root.glob(args.glob))
               if (p / "modalities").is_dir()]
    print(f"Found {len(ep_dirs)} episodes under {root}  (workers={args.workers})")

    ok = skip = fail = 0
    with ProcessPoolExecutor(max_workers=args.workers) as ex:
        futs = {ex.submit(process_episode, str(ep), args.video_name, args.overwrite): ep
                for ep in ep_dirs}
        for fut in as_completed(futs):
            ep = futs[fut]
            try:
                name, status = fut.result()
                print(f"[{name}] {status}")
                if status.startswith("ok"):
                    ok += 1
                else:
                    skip += 1
            except Exception:
                fail += 1
                print(f"[{ep.name}] FAILED\n{traceback.format_exc()}")

    print(f"\nDone. ok={ok}  skip={skip}  fail={fail}")


if __name__ == "__main__":
    main()