# -*- coding: utf-8 -*-
"""按《视频抽帧与字幕规范》严格重抽关键帧。

与 v1 的差异：
- 字幕区 scene 检测严格使用底部 15%（B 站 1920x1080 -> 162px；按比例适配其他分辨率）。
- 新增字幕区低阈值 0.05 作为安全网，防止缓变换句被漏检。
- 输出 frame_times.json，记录每帧来自哪种检测（便于核对）。
- 保留首帧 0s、尾帧 DUR-0.3s、>5s 长间隔补中点的规范步骤。

用法：
  python extract_frames_v2.py --dir <视频目录> --out <帧输出目录> --ff <ffmpeg bin 目录>
"""
import argparse, glob, json, os, re, subprocess, sys


def run(cmd):
    return subprocess.run(cmd, capture_output=True, text=True, encoding="utf-8", errors="replace")


def probe(video, ffprobe):
    r = run([ffprobe, "-v", "error", "-select_streams", "v:0",
             "-show_entries", "stream=width,height", "-show_entries", "format=duration",
             "-of", "default=nk=1:nw=1", video])
    vals = [x.strip() for x in r.stdout.split("\n") if x.strip()]
    return int(vals[0]), int(vals[1]), float(vals[2])


def scene_times(video, vf, ffmpeg):
    """跑一次 scene 检测并返回所有 pts_time（秒）。"""
    r = run([ffmpeg, "-i", video, "-vf", vf + ",showinfo", "-vsync", "vfr", "-f", "null", "-"])
    out = set()
    for line in r.stderr.splitlines():
        m = re.search(r"pts_time:([0-9.]+)", line)
        if m:
            out.add(round(float(m.group(1)), 4))
    return sorted(out)


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--dir", required=True, help="含 .mp4 的目录")
    ap.add_argument("--out", required=True, help="帧输出目录")
    ap.add_argument("--ff", default=r"D:\ComfyUI_windows\ffmpeg-8.0.1-full_build\bin",
                    help="ffmpeg/ffprobe 所在 bin 目录")
    args = ap.parse_args()

    ffmpeg = os.path.join(args.ff, "ffmpeg.exe")
    ffprobe = os.path.join(args.ff, "ffprobe.exe")

    vids = sorted(glob.glob(os.path.join(args.dir, "*.mp4")))
    if not vids:
        print("✗ 目录里没有 .mp4：%s" % args.dir)
        sys.exit(1)
    video = vids[0]
    print("视频：%s" % video)

    w, h, dur = probe(video, ffprobe)
    print("分辨率=%dx%d  时长=%.3fs" % (w, h, dur))

    # 字幕区用于 scene 检测：底部 15%（规范参考值）
    sub_h = max(1, int(round(h * 0.15)))
    sub_y = h - sub_h
    crop_vf = "crop=%d:%d:0:%d" % (w, sub_h, sub_y)
    print("字幕区裁剪（检测用）：%s" % crop_vf)

    # ① 画面变化
    t_scene_03 = scene_times(video, "select='gt(scene,0.3)'", ffmpeg)
    # ② 字幕变化（规范核心）
    t_sub_01 = scene_times(video, "%s,select='gt(scene,0.1)'" % crop_vf, ffmpeg)
    # ③ 低阈值补漏（规范标准步骤）
    t_scene_005 = scene_times(video, "select='gt(scene,0.05)'", ffmpeg)
    # 额外安全网：字幕区低阈值，捕获缓变换句
    t_sub_005 = scene_times(video, "%s,select='gt(scene,0.05)'" % crop_vf, ffmpeg)

    print("scene 0.3=%d  字幕区 0.1=%d  全局 0.05=%d  字幕区 0.05=%d"
          % (len(t_scene_03), len(t_sub_01), len(t_scene_005), len(t_sub_005)))

    # 合并 + 首尾
    all_times = sorted(set(t_scene_03) | set(t_sub_01) | set(t_scene_005) | set(t_sub_005))
    all_times.append(0.0)
    all_times.append(max(0.0, round(dur - 0.3, 3)))
    all_times = sorted(set(all_times))

    # >5s 长间隔补中点
    filled = []
    prev = None
    for t in all_times:
        if prev is not None and (t - prev) > 5:
            filled.append(round((t + prev) / 2, 3))
        filled.append(t)
        prev = t
    times = sorted(set(filled))
    print("最终帧数（含首尾+补帧）：%d" % len(times))

    # 记录来源（用于调试/核对）
    source_map = {}
    for t in times:
        tags = []
        if t == 0.0: tags.append("start")
        if abs(t - (dur - 0.3)) < 0.001: tags.append("end")
        if t in t_scene_03: tags.append("scene03")
        if t in t_sub_01: tags.append("sub01")
        if t in t_scene_005: tags.append("scene005")
        if t in t_sub_005: tags.append("sub005")
        if "start" not in tags and "end" not in tags and all(t not in s for s in (t_scene_03, t_sub_01, t_scene_005, t_sub_005)):
            tags.append("fill")
        source_map["%.3f" % t] = tags

    os.makedirs(args.out, exist_ok=True)
    for i, t in enumerate(times, 1):
        tt = t if t <= dur else max(0.0, dur - 0.3)
        outp = os.path.join(args.out, "c_%04d_%.3fs.jpg" % (i, t))
        r = run([ffmpeg, "-y", "-ss", "%.3f" % tt, "-i", video,
                 "-frames:v", "1", "-q:v", "2", outp])
        if r.returncode != 0:
            print("  ✗ 帧 %d 失败：%s" % (i, r.stderr[-200:]))

    meta = {
        "video": video, "width": w, "height": h, "duration": dur,
        "subtitle_crop": crop_vf,
        "counts": {
            "scene_0.3": len(t_scene_03), "sub_0.1": len(t_sub_01),
            "scene_0.05": len(t_scene_005), "sub_0.05": len(t_sub_005),
            "final": len(times)
        },
        "frame_times": source_map
    }
    with open(os.path.join(args.out, "frame_times.json"), "w", encoding="utf-8", newline="\n") as f:
        json.dump(meta, f, ensure_ascii=False, indent=2)

    print("✓ 抽帧完成 → %s" % args.out)


if __name__ == "__main__":
    main()