Hi all!
The technology showcased here is super interesting. Our team does of course review recent papers, and we did also review this one. Unfortunately, this tech is not as perfect as it seems. The major issues would be that it handles only one GI bounce correctly, and that it works fine only on a small set of pre-selected scenes. We will definitely keep improving our in-house caustics solver, though.
Please do send us any interesting papers your find, on caustics and other topics too!