Computer Graphics Laboratory ETH Zurich


Deep Compositional Denoising on Frame Sequences

X. Zhang, G. Roethlin, M. Manzi, M. Gross, M. Papas

Eurographics Symposium on Rendering (EGSR) (Delft, The Netherlands, June 28-30, 2023), Industry Track


We propose a novel extension for processing frame sequences that provides context from neighboring frames while avoiding redundant computation and a novel quantization approach that reduces the memory overhead without introducing artifacts. Our method achieves a significant improvement in denoising quality over KPAL and ameliorates temporal denoising artifacts from the single-frame approach. This enables the use of the compositional denoiser in production settings.


Download Paper
Download Paper