DIST: Rendering Deep Implicit Signed Distance Function with Differentiable Sphere Tracing

TL;DR

This paper introduces a differentiable sphere tracing algorithm for efficient rendering of deep implicit signed distance functions, enabling accurate 3D shape reconstruction from limited inputs with robust inverse optimization.

Contribution

It presents a novel differentiable rendering method that optimizes both forward and backward passes for efficient neural implicit shape reconstruction.

Findings

Effective 3D shape reconstruction from sparse data

Robustness against noise in shape prediction

High generalization capability of the method

Abstract

We propose a differentiable sphere tracing algorithm to bridge the gap between inverse graphics methods and the recently proposed deep learning based implicit signed distance function. Due to the nature of the implicit function, the rendering process requires tremendous function queries, which is particularly problematic when the function is represented as a neural network. We optimize both the forward and backward passes of our rendering layer to make it run efficiently with affordable memory consumption on a commodity graphics card. Our rendering method is fully differentiable such that losses can be directly computed on the rendered 2D observations, and the gradients can be propagated backwards to optimize the 3D geometry. We show that our rendering method can effectively reconstruct accurate 3D shapes from various inputs, such as sparse depth and multi-view images, through inverse optimization. With the geometry based reasoning, our 3D shape prediction methods show excellent generalization capability and robustness against various noises.