Skip to content

[KDD'2026] "VideoRAG: Chat with Your Videos"

3.5 KB
Raw Blame
import os
import time
import shutil
import numpy as np
from tqdm import tqdm
from moviepy.video import fx as vfx
from moviepy.video.io.VideoFileClip import VideoFileClip
from .._utils import logger

def split_video(
    video_path,
    working_dir,
    segment_length,
    num_frames_per_segment,
    audio_output_format='mp3',
    audio_sample_rate=16000,  # Default 16kHz for speech recognition
):  
    unique_timestamp = str(int(time.time() * 1000))
    video_name = os.path.basename(video_path).split('.')[0]
    video_segment_cache_path = os.path.join(working_dir, '_cache', video_name)
    if os.path.exists(video_segment_cache_path):
        shutil.rmtree(video_segment_cache_path)
    os.makedirs(video_segment_cache_path, exist_ok=False)
    
    segment_index = 0
    segment_index2name, segment_times_info = {}, {}
    with VideoFileClip(video_path) as video:
    
        total_video_length = int(video.duration)
        start_times = list(range(0, total_video_length, segment_length))
        # if the last segment is shorter than 5 seconds, we merged it to the last segment
        if len(start_times) > 1 and (total_video_length - start_times[-1]) < 5:
            start_times = start_times[:-1]
        
        for start in tqdm(start_times, desc=f"Spliting Video {video_name}"):
            if start != start_times[-1]:
                end = min(start + segment_length, total_video_length)
            else:
                end = total_video_length
            
            subvideo = video.subclip(start, end)
            subvideo_length = subvideo.duration
            frame_times = np.linspace(0, subvideo_length, num_frames_per_segment, endpoint=False)
            frame_times += start
            
            segment_index2name[f"{segment_index}"] = f"{unique_timestamp}-{segment_index}-{start}-{end}"
            segment_times_info[f"{segment_index}"] = {"frame_times": frame_times, "timestamp": (start, end)}
            
            # save audio
            audio_file_base_name = segment_index2name[f"{segment_index}"]
            audio_file = f'{audio_file_base_name}.{audio_output_format}'
            subaudio = subvideo.audio
            # Convert to mono and set sample rate using ffmpeg parameters
            subaudio.write_audiofile(
                os.path.join(video_segment_cache_path, audio_file), 
                codec='mp3',
                fps=audio_sample_rate,  # Set sample rate
                ffmpeg_params=['-ac', '1'],  # Force mono (1 audio channel)
                verbose=False, 
                logger=None
            )
            
            segment_index += 1

    return segment_index2name, segment_times_info

def saving_video_segments(
    video_name,
    video_path,
    working_dir,
    segment_index2name,
    segment_times_info,
    video_output_format='mp4',
):
    try:
        with VideoFileClip(video_path) as video:
            video_segment_cache_path = os.path.join(working_dir, '_cache', video_name)
            for index in tqdm(segment_index2name, desc=f"Saving Video Segments {video_name}"):
                start, end = segment_times_info[index]["timestamp"][0], segment_times_info[index]["timestamp"][1]
                video_file = f'{segment_index2name[index]}.{video_output_format}'
                subvideo = video.subclip(start, end)
                subvideo.write_videofile(os.path.join(video_segment_cache_path, video_file), codec='libx264', verbose=False, logger=None)
    except Exception as e:
        logger.error(f"Error in saving_video_segments:\n {str(e)}")
        raise RuntimeError