Thriving 
RS 
PY zensim research: training, evaluation and experiment records. The published zensim crate lives in imazen/zensim.
https://imazen.github.io/zensimx/ ↗ HEALTH ? 90 / Healthy
Thriving RS PY // readme
zensim
This is zensimx, the zensim research repository (renamed from imazen/zensim on 2026-10-28): training, evaluation, experiment records and the research API. The published
zensimcrate is developed and released from imazen/zensim, which continues from zensim 0.2.7. Thezensimandzensim-regresspackages here arepublish = false.
A perceptual image quality score for codec quality targeting: users choose one score, and encoders choose the settings to reach it.
Built on the same psychovisual foundations as SSIMULACRA2 and butteraugli — multi-scale SSIM, edge artifacts, detail loss, and high-frequency features in XYB color space — but with trained weights, fused SIMD kernels, and multi-threaded computation.
Current checkout (2026-10-19): the user-facing control is one target score.
codec_target() returns B / zensim-b: diff+frag-820 (Rev5); D is an explicit…