Self-Supervised Post-Correction for Monte Carlo Denoising
Event Type
Technical Paper
Interest Areas
Research & Education
Presentation Types
Registration Categories
Full Conference Supporter
Full Conference
Virtual Conference Supporter
Virtual Conference
Exhibitor Additional Full Conference
Exhibitor Full Conference
DescriptionWe propose a novel post-correction neural network that can correct the outputs of learning-based denoising techniques. This paper derives a self-supervised loss that can guide our post-correction neural network to optimize its parameters on the fly, using only a test image pair (i.e., a noisy image and its denoised output).