Forget Superresolution, Sample Adaptively (when Path Tracing) CCS: Computing methodologies Neural networks CCS: Computing methodologies Computer graphics
Martin Bálint, Corentin Salaün, Hans-Peter Seidel, Karol Myszkowski · 2026
Read Forget Superresolution, Sample Adaptively (when Path Tracing) CCS: Computing methodologies Neural networks CCS: Computing methodologies Computer graphics by Martin Bálint, Corentin Salaün, Hans-Peter Seidel, Karol Myszkowski (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-08642.txt.
Description
Real-time path tracing now runs below one sample per pixel, where super-resolution smears detail uniformly. Martin Bálint and three co-authors propose adaptive sampling instead, with an end-to-end pipeline built for the sub-one-sample regime. Their arXiv paper targets the sparse budgets that modern resolution and frame rates demand.
A stochastic sample-placement formulation allows gradient estimation despite discrete choices, stabilizing neural sampler training. Tonemapping-aware training with perceptual loss avoids wasting samples on low-impact regions, and a gather-based pyramidal denoiser with learnable albedo handling cleans the result. Tests show consistent gains over uniform sparse sampling, especially on highlights and shadow edges. Readers get a case for sampling smarter, not just upscaling.
Catalog record
| CREATOR | Martin Bálint, Corentin Salaün, Hans-Peter Seidel, Karol Myszkowski |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 76 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-08642