Approximate Differentiable Rendering with Algebraic Surfaces

TL;DR

This paper introduces Fuzzy Metaballs, an approximate differentiable renderer that offers faster computation and smooth outputs for shape rendering, enabling effective vision tasks like pose estimation and shape from silhouette.

Contribution

It presents a novel, efficient differentiable renderer based on algebraic surfaces that balances speed and quality, suitable for practical vision applications.

Findings

Fuzzy Metaballs renderer is 5x faster in forward passes and 30x faster in backward passes than mesh-based methods.

The method produces smooth, well-defined depth maps and silhouettes.

It achieves competitive results in pose estimation and shape reconstruction without surrogate losses.

Abstract

Differentiable renderers provide a direct mathematical link between an object's 3D representation and images of that object. In this work, we develop an approximate differentiable renderer for a compact, interpretable representation, which we call Fuzzy Metaballs. Our approximate renderer focuses on rendering shapes via depth maps and silhouettes. It sacrifices fidelity for utility, producing fast runtimes and high-quality gradient information that can be used to solve vision tasks. Compared to mesh-based differentiable renderers, our method has forward passes that are 5x faster and backwards passes that are 30x faster. The depth maps and silhouette images generated by our method are smooth and defined everywhere. In our evaluation of differentiable renderers for pose estimation, we show that our method is the only one comparable to classic techniques. In shape from silhouette, our method performs well using only gradient descent and a per-pixel loss, without any surrogate losses or regularization. These reconstructions work well even on natural video sequences with segmentation artifacts. Project page: https://leonidk.github.io/fuzzy-metaballs