Closed AminMemariani closed 4 months ago
Hi, The code in deep-shadow needs binary shadow maps. I suppose you could try training a model to extract shadow maps from glossy objects (assuming it's a simpler problem than predicting a dense depth/normal maps), and from there, run deep-shadow. You can try achieving this by creating a synthetic dataset which has tuples of glossy objects and their corresponding shadow maps, and then use that to train a model.
I can't ensure this will work, but you can try. If you are able to train a model for the first step, then deep-shadow should work for you.
Hi, The code in deep-shadow needs binary shadow maps. I suppose you could try training a model to extract shadow maps from glossy objects (assuming it's a simpler problem than predicting a dense depth/normal maps), and from there, run deep-shadow. You can try achieving this by creating a synthetic dataset which has tuples of glossy objects and their corresponding shadow maps, and then use that to train a model.
I can't ensure this will work, but you can try. If you are able to train a model for the first step, then deep-shadow should work for you.
Thank you
In my case, I want to reconstruct 3D objects from 2D images. The 2d images are captured from convex and glossy objects. Do you think this repo can still be useful in that case?