Fast SSIMULACRA2 Implementation in Zig
a day ago
- #Zig
- #SSIMULACRA2
- #Performance
- Fast SSIMULACRA2 derivative implementation in Zig, named fssimu2.
- Supports sRGB PNG, PAM, JPEG, WebP, or AVIF input formats.
- Performance tested on Intel Core i7 13700k, showing up to 23% faster and ~40% less memory usage compared to reference implementation.
- Includes options for JSON output and error map saving.
- Conformance to reference SSIMULACRA2 tested with validate.py, showing high correlation (PCC: 0.999700).
- Compilation requires Zig (version 0.15.1), libjpeg-turbo, libwebp, and libavif.
- Provides a C-compatible ABI with ssimu2.h include file for SSIMULACRA2 score computation.
- Example C program provided in c_abi_example/ directory.
- Project is under Apache 2.0 license, with third-party code from libspng, libminiz, and vapoursynth-zip.