The most successful Monte Carlo denoising calculates the filter weights in a heuristic manner. Learning Based Filtering approaches this via the use of a neural network to automatically learn the complex relationship between the noisy scene and the ideal filter parameters, which is not possible via the heuristic rules. The automatic learning process will bring the next level of efficiency into the filtering of Monte Carlo images in a train and forget method such that the filter weights need not be calculated for each noisy image being filtered.
The holy grail of filtering is to be able to filter the noise of an image in real time. This is the final frontier in filtering of noisy images. innoBright is working together with leading researchers in the field to crack this final frontier.