Fitting Neural Textures and PBR Material Maps with ES (No Backprop)
ibobev
13 points
3 comments
September 14, 2026
Related Discussions
Found 5 related stories in 110.9ms across 10,991 title embeddings via pgvector HNSW
- Anatomy of a Texture Agentlien · 86 pts · September 16, 2026 · 44% similar
- Dust: Pretraining Transformers Without Backpropagation E-Reverance · 147 pts · October 05, 2026 · 41% similar
- Show HN: MultiMatte, a Promptable Image Background Removal Model snyy · 45 pts · September 10, 2026 · 40% similar
- Retrospectively Reverse-Engineering Apple's Neural Engine zdw · 225 pts · September 12, 2026 · 40% similar
- A Texture Lookup Approach to Bézier Curve Evaluation on the GPU (JCGT) ibobev · 37 pts · July 29, 2026 · 39% similar
Discussion Highlights (2 comments)
yhjc2692
Extremely cool stuff. I’m excited for neural approaches to further improve the bitrate frontier of “visually lossless” compression
cl3misch
> no backprop, no autodiff, no training framework OP says it like it's a good thing, but doesn't explain why. To not rely on Pytorch? What's the benefit?