AMD paper: a single generalized MLP upscales textures on-the-fly at render time, saving VRAM and disk
ssh4net · x · 2026-11-06
AMD researchers Shin Fujieda and Takahiro Harada propose 'Patch-Based Generalized Neural Texture Upscaling': a lightweight MLP (3x3 texel input, 2x2 output, with residual blocks and an analytic-upscaling residual connection) trained on diverse texture patches and integrated into the texture sampling stage to upscale low-res textures on-the-fly at render time. Because the model is generalized rather than overfit per asset, it applies to any existing scene without changing textures or the asset pipeline. Benefits: smaller disk footprint, no high-res textures in VRAM, one shared weight set ensures non-divergent GPU execution. Evaluations (FLIP on the HANGER scene, LPIPS on Poly Haven textures) beat bilinear, bicubic, and PFT.
More from Research
- Lecture derives R-GCN's 4.75M-parameter bill and a 14.7× basis-decomposition fix — JeremyCMorgan · 2026-11-06
- Random steering vectors turn LLMs into teapots, integers and 2019 movies — Sauers_ · 2026-11-06
- Epoch AI review: OSWorld tasks are simple, unstable, and barely test GUI use — gleech · 2026-11-06
- One random vector in 240 makes Qwen3-30B believe it's a cat via MoE steering — Sauers_ · 2026-11-06
- Steering a single MoE expert at strength +1.25 rewires Qwen3-30B's persona — Sauers_ · 2026-11-06
- Harvard-led team's MFM memory system lets frozen-weight agents keep learning in Minecraft — du_yilun · 2026-11-06