How far off is this from a real-world application? Quoting the original paper:
[UWSL]Our method integrates learning-based approaches with[/UWSL][UWSL] conventional real-time rendering pipelines. We expect our[/UWSL][UWSL] method to continue to benefit future graphics pipelines and[/UWSL][UWSL] to be compatible with real-time ray tracing. Inference with [/UWSL][UWSL]our approach in its current unoptimized implementation[/UWSL][UWSL] takes half a second on a Geforce RTX 3090 GPU. Since G-[/UWSL][UWSL]buffers that are used as input are produced natively on the[/UWSL][UWSL] GPU, our method could be integrated more deeply into game[/UWSL][UWSL] engines, increasing efficiency and possibly further advancing[/UWSL][UWSL] the level of realism.[/UWSL]