Video Dataset Preparation utilities

Split_scenes.py

I recommend using the split_scenes.py file from the LTX-Video-Trainer repo for this purpose.

It is a Python utility for automatically detecting and splitting videos into individual scenes based on visual content changes.

Frame_Bucketeer.py

A Python utility that analyzes MP4 files by frame count and organizes them into appropriate buckets.

Key Features

• Analyzes MP4 files to determine their exact frame count using OpenCV
• Organizes videos into user-defined frame count buckets (e.g., 30, 60, 120 frames)
• Creates only the bucket folders that will actually contain files
• Places each video in the bucket with the closest smaller-or-equal frame count
• Provides detailed progress feedback and summary statistics

Requirements

• Python 3.6+
• OpenCV (pip install opencv-python)

Usage

python bucketeer.py /path/to/videos --buckets 30,60,120,300

FPS_changer.py

A simple Python script to change the FPS (frames per second) of all videos in a folder.

Key Features

• This tool changes the frame rate but keeps the same video duration
• It processes all videos in the input folder and saves them to the output folder
• Original videos are left untouched
• Audio is preserved without re-encoding

Requirements

• Python 3.6 or higher
• FFmpeg (must be installed separately and available in your PATH)

Usage

python fps-changer.py --input /path/to/videos --output /path/to/output --fps 16

video_res_fps_analyzer.py

A simple Python script to analyze video files and display their resolution, aspect ratio, frame count, and FPS information.

Key Features

Analyzes all video files in a folder and displays comprehensive information Shows resolution, aspect ratio, total frame count, and FPS for each video Supports recursive scanning of subdirectories Works with a wide range of video formats (MP4, AVI, MKV, MOV, and many more) Fast processing using FFprobe Debug mode for troubleshooting file detection issues

Requirements

• Python 3.6 or higher
• FFmpeg with FFprobe (must be installed separately and available in your PATH)

Installation

Install FFmpeg:

• Windows: Download from https://ffmpeg.org/download.html
• macOS: brew install ffmpeg
• Linux: sudo apt install ffmpeg (Ubuntu/Debian) or sudo yum install ffmpeg (CentOS/RHEL)

Verify FFprobe is working: ffprobe -version

Usage

Basic usage:

python video_res_fps_analyzer.py /path/to/videos

Command Line Options

• folder - Input folder path (required)
• -r, --recursive - Search recursively in subdirectories
• --no-duration - Skip frame count and FPS information for faster processing
• --debug - Enable debug output to troubleshoot file detection issues
• -h, --help - Show help message

ar_normalizer.py

A Python utility that normalizes video aspect ratios using intelligent cropping while preserving maximum quality.

Key Features

• Normalizes all videos in a folder to a target aspect ratio (default 16:9)
• Uses intelligent cropping to maintain content without black bars or distortion
• Preserves original quality when videos already have the correct aspect ratio (simple copy)
• Maintains highest possible quality for processed videos (CRF 18, audio copy)
• Automatically calculates optimal output dimensions based on input video resolution
• Supports all common video formats (MP4, MOV, AVI, MKV, M4V, WMV, FLV, WebM)

Requirements

• Python 3.6+
• FFmpeg (must be installed and available in PATH)
• FFprobe (usually included with FFmpeg)

Usage

Basic usage (16:9 aspect ratio):

python ar_normalizer.py /path/to/input /path/to/output

Custom aspect ratio:

python ar_normalizer.py /path/to/input /path/to/output --aspect-ratio 1.33

Fixed width or height:

# All videos will have 1920px width, height calculated from aspect ratio
python ar_normalizer.py /path/to/input /path/to/output --aspect-ratio 1.78 --width 1920

# All videos will have 1080px height, width calculated from aspect ratio  
python ar_normalizer.py /path/to/input /path/to/output --aspect-ratio 1.78 --height 1080

How It Works

Smart Processing:

• Videos with correct aspect ratio (±0.01 tolerance) are simply copied without re-encoding
• Videos needing correction are cropped from the center and scaled to target dimensions
• Audio is always preserved without re-encoding to maintain quality

Cropping Logic:

• Square videos (1:1) → 16:9: Crops top and bottom, maintains width
• Portrait videos (9:16) → 16:9: Crops top and bottom significantly
• Wide videos (2:1) → 16:9: Crops left and right sides
• Always crops from the center to preserve the most important content

Common Aspect Ratios

• 1.78 (16:9) - Standard widescreen, YouTube, most displays
• 1.33 (4:3) - Classic TV, older content
• 1.0 (1:1) - Square format, Instagram posts
• 0.56 (9:16) - Vertical format, TikTok, Instagram Stories
• 2.35 (21:9) - Cinematic widescreen