3 ms·
It's likely mostly for denoising, I highly doubt it can have enough power to perform both denoising and antialiasing. The ray tracing presented would be samplin
by xxs 8y ago
It's likely mostly for denoising, I highly doubt it can have enough power to perform both denoising and antialiasing. The ray tracing presented would be sampling and then use the model to fill the gaps.
- mattnewport 8y agoYou're effectively training the network for both when training for denoising as your ground truth images are both denoised and anti aliased. With ray tracing both problems are brute forced by the same approach of throwing more samples per pixel at the problem.