"I am very excited about this work. I believe it significantly advances the field of neural rendering and will be very influental for future research. The scenes covered in the paper are sufficiently hard to render in real-time and cover many failure cases that are typical for many-light methods. The results are stunningly temporally stable and look great, even when compared with a path tracer and denoising. With the current AI gold rush, it is to be expected that machine learning accelerators will play increasingly bigger roles. This paper will be an highly influential work in bridging the worlds of real-time rendering and machine learning."

"This is a neat way of exploiting a correspondence between a rendering problems and modern ML technique. I find it inspiring. The idea may found adoption in other transport setups."