视频调整帧率、分辨率+音画同步

# python data_utils/pre_video/multi_fps_crop_sync.py

import cv2
import os
from tqdm import tqdm
import subprocess

# 加载人脸检测模型
face_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')

def contains_face(frame):
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    faces = face_cascade.detectMultiScale(gray, 1.3, 5)
    return len(faces) > 0

def crop_center(frame, crop_width, crop_height):
    height, width = frame.shape[:2]
    start_x = width//2 - crop_width//2
    start_y = height//2 - crop_height//2
    return frame[start_y:start_y+crop_height, start_x:start_x+crop_width]

def extract_audio(input_path, audio_path):
    subprocess.run(['ffmpeg', '-y', '-i', input_path, '-vn', '-acodec', 'copy', audio_path])

def merge_video_audio(video_path, audio_path, output_path):
    # 使用ametadata滤镜将音频的时间戳与视频流的时间戳对齐
    subprocess.run([
        'ffmpeg', '-y', '-i', video_path,