Using a Waffle Iron for Automotive Point Cloud Semantic Segmentation

TL;DR

This paper introduces WaffleIron, a novel 3D backbone for semantic segmentation of point clouds that achieves competitive performance without using sparse 3D convolutions, making it suitable for hardware lacking such capabilities.

Contribution

The paper presents WaffleIron, a dense 2D and MLP-based 3D backbone that rivals state-of-the-art sparse convolution methods in semantic segmentation tasks.

Findings

WaffleIron achieves high performance on SemanticKITTI and nuScenes datasets.

It offers an alternative approach suitable for hardware without sparse convolution support.

The method simplifies the architecture by avoiding sparse 3D convolutions.

Abstract

Semantic segmentation of point clouds in autonomous driving datasets requires techniques that can process large numbers of points efficiently. Sparse 3D convolutions have become the de-facto tools to construct deep neural networks for this task: they exploit point cloud sparsity to reduce the memory and computational loads and are at the core of today's best methods. In this paper, we propose an alternative method that reaches the level of state-of-the-art methods without requiring sparse convolutions. We actually show that such level of performance is achievable by relying on tools a priori unfit for large scale and high-performing 3D perception. In particular, we propose a novel 3D backbone, WaffleIron, made almost exclusively of MLPs and dense 2D convolutions and present how to train it to reach high performance on SemanticKITTI and nuScenes. We believe that WaffleIron is a compelling alternative to backbones using sparse 3D convolutions, especially in frameworks and on hardware where those convolutions are not readily available.