Converting objects in a 2D image to 3D meshes

I have been using Neural Mesh Renderer multiperson/neural_renderer at master · JiangWenPL/multiperson · GitHub for converting 2D image to 3D mesh for an object (e.g. teapot as shown in the github example).

Is this the best package for this purpose in PyTorch or are there better and more precise candidates for this purpose? (given it’s a bit outdated by now – CVPR 2018)

Also, can PyTorch3D do the same thing? I was not able to find an end-to-end example that can do so. If you could link me with one would be very fantastic.

P.S.: I also see this GitHub - ShichenLiu/SoftRas: Project page of paper "Soft Rasterizer: A Differentiable Renderer for Image-based 3D Reasoning" (haven’t fully set it up) but PyTorch3D first author mentions they are inspired by this paper - assuming they perform better?

1 Like