NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center
Paper ID:5098
Title:Learning to Predict 3D Objects with an Interpolation-based Differentiable Renderer


		
This paper received high-variance reviews, with some in favor, some borderline, and one against. Ultimately, the decision was made to accept this paper: a differentiable rasterizer that supports the full set of standard rendering features (lighting, texture, etc) in one package is a valuable contribution. The committee does have one reservation. The main contribution of the paper is a differentiable form of rendering based on Barycentric interpolation within triangles. The authors claim this as a novel algorithm, but this is actually a standard procedure used by virtually all rasterization-based renderers. The authors should tone down their claims of algorithmic novelty and explicitly acknowledge the strong connection to classic rasterization.