Open woduck opened 3 weeks ago
Hi @woduck. Sure, one could also use a single image as input and still get natural envmaps. We show the single input result in Fig. 1 and sec. 5.6 in our paper. However, the assumption is a well-initialized geometry is given, so the geometry is either reconstructed from multiview or 3D scan.
In your DPI paper, it is mentioned that it is possible to generate the envmap and pbrmaps with a single image input. However, upon reviewing the actual code, I noticed that it selects 60 images randomly as input.
I would like to know if you have conducted any experiments with a single image input. If so, what were the results?