isl-org/PhotorealismEnhancement

Dataset release

ydblg opened this issue · 2 comments

ydblg commented

Hello. First, congratulations on the great work and the amazing results. Will the dataset used for training be released ? If yes, do you have a time window ?

Same question here, could you please release the training data?

Hi,
thanks for the kind words. As it appears now, we will unfortunately not be able to release the full dataset, i.e. all of the G-buffers due to legal issues. This should not keep you from training your own models with the code, though -- several other projects have demonstrated how G-buffers can be captured from games and have released code for doing so. Also, please keep in mind that the G-buffers are only one component in the whole pipeline and strong results can be achieved even when just using rendered images without any G-buffers (as demonstrated in our ablation study in the paper).