Wavelet Bench

July 7, 2025 ยท View on GitHub

A video encoding benchmarking tool for wavelet-based video codecs.

Overview

Wavelet Bench is a benchmarking toolkit designed to evaluate the performance of various video encoders, with a focus on wavelet-based codecs. It provides tools to generate comprehensive metrics including PSNR, SSIM, and XPSNR, allowing for detailed analysis and comparison of different encoding algorithms.

TODO

  • DSV2 support
  • Dirac support
  • x264 support
  • Snow support

Features

  • Support for multiple wavelet-based encoders, including:
    • Snow (FFmpeg's wavelet codec)
    • Dirac (BBC's open-source wavelet codec)
    • DSV2 (custom wavelet encoder)
    • X264 (for comparison)
  • Automated quality testing across multiple quality levels
  • Detailed metrics collection:
    • PSNR (Peak Signal-to-Noise Ratio)
    • SSIM (Structural Similarity Index)
    • XPSNR (Extended PSNR for Y, U, V channels)
    • W-XPSNR (Weighted XPSNR)
  • CSV output for easy analysis and visualization
  • Performance tracking (encoding time and file size)

Requirements

  • Python 3.13.2+
  • FFmpeg (with XPSNR support)
  • X264 (optional, for comparison testing)
  • Other codec implementations as needed

Installation

Dependencies

  • uv, a Python project manager
  • FFmpeg >= 7.1 (for XPSNR calculations)

Install Steps

  1. Clone the repository:

    git clone https://github.com/username/wavelet-bench.git
cd wavelet-bench

  2. Mark the scripts as executable:

    cd scripts/
chmod a+x encode.py
chmod a+x stats.py

  3. Run the script of your choice:

    ./stats.py --help

Usage

Single Encode Test

To test a single video encode and see detailed metrics:

./encode.py -i /path/to/video.mp4 -q 30 x264 -- --preset medium

Options:

  • -i, --input: Path to source video file (required)
  • -q, --quality: Quality/CRF value for the encoder (required)
  • encoder: Which encoder to use (snow, dsv2, dirac, x264)
  • -b, --out: Output video filename (optional)
  • encoder_args: Additional encoder arguments (pass after --)

Batch Testing

For comprehensive testing across multiple quality settings and videos:

./stats.py -i video1.mp4 video2.mp4 -q "20 30 40 50" x264 -o results.csv -- --preset medium

Options:

  • -i, --inputs: One or more source video files (required)
  • -q, --quality: Space-separated list of quality values to test (required)
  • encoder: Which encoder to use (snow, dsv2, dirac, x264)
  • -o, --output: Output CSV file path (required)
  • -k, --keep: Keep encoded video files (default is to delete)
  • encoder_args: Additional encoder arguments (pass after --)

License

See License for more info.