... stumbled upon the following study which could potentially be the secret ingredientkintuX wrote: ↑Tue Mar 24, 2020 9:23 pmAbout Clarisse IFX Clamping:It's been developed in-house and no papers or other technicalities have been revealed yet.Since Clarisse 4.0 SP5, the path tracer offers a novel method to reduce considerably fireflies in renders. For more information please refer to the Fireflies Filtering section.
Antialasing Appearance in Global Illumination Renderers (2015)
Video on YTIn Physically Based Rendering (PBR) detailed objects can show aliasing once rendered. Such aliasing can be avoided using super sampling, but this increases dramatically the rendering cost. Instead, it is preferable to antialias the appearance of objects. Bandlimiting appearance requires two elements: an appearance model compatible with bandlimiting (such as LEAN mapping, LEADR mapping, etc), and the ability to determine the appearance bandlimit before the evaluation of this appearance model.
We provide the first theoretical analysis of how to perform antialiasing in modern path tracing engines (hence supporting Bidirectional Path Tracing) and we provide a practical solution to determine the appearance bandlimit. While antialiased surface appearance models are now widely used in production, not much work has been done on the antialiasing kernel since the unidirectional works of Igehy  and Suykens and Willems . Our work is also motivated by the increased adoption of bidirectional path tracers in production renderers (Weta’s Manuka, Pixar’s PRman). The core of our theoretical contribution is to show derive filtering footprints accounting for both eye path and light path’s frequency content. Practically, we use covariance tracing to efficiently track down an approximation of the filtering footprint. Furthermore, we show that covariance tracing can replace ray differentials as it removes its limitations (specular interaction) while maintaining backward compatibility (to easy integration). We believe that our method should in fine replace completely ray differentials (removing an unnecessary conversion cost).
In other news, OIDN was updated to v1.2:
- Added neural network training code
- Added support for specifying user-trained models at runtime
- Slightly improved denoising quality (e.g. less ringing artifacts, less blurriness in some cases)
- Improved denoising speed by about 7-38% (mostly depending on the compiler)
- Added OIDN_STATIC_RUNTIME CMake option (for Windows only)
- Added support for OpenImageIO to the example apps (disabled by default)
- Added check for minimum supported TBB version
- Find debug versions of TBB
- Added testing