Point Cloud Semantic Segmentation with Sparse and Inhomogeneous Annotations

TL;DR

This paper introduces AADNet, a method for point cloud semantic segmentation that effectively handles inhomogeneous sparse annotations by analyzing their impact and employing adaptive strategies for robust learning.

Contribution

The work proposes a novel analysis of annotation sparsity and inhomogeneity, and develops AADNet with label-aware downsampling and gradient calibration for improved weakly supervised segmentation.

Findings

Achieves significant performance improvements across various label rates.

Effectively mitigates gradient bias caused by non-uniform annotations.

Operates without prior restrictions or additional information.

Abstract

Utilizing uniformly distributed sparse annotations, weakly supervised learning alleviates the heavy reliance on fine-grained annotations in point cloud semantic segmentation tasks. However, few works discuss the inhomogeneity of sparse annotations, albeit it is common in real-world scenarios. Therefore, this work introduces the probability density function into the gradient sampling approximation method to qualitatively analyze the impact of annotation sparsity and inhomogeneity under weakly supervised learning. Based on our analysis, we propose an Adaptive Annotation Distribution Network (AADNet) capable of robust learning on arbitrarily distributed sparse annotations. Specifically, we propose a label-aware point cloud downsampling strategy to increase the proportion of annotations involved in the training stage. Furthermore, we design the multiplicative dynamic entropy as the gradient calibration function to mitigate the gradient bias caused by non-uniformly distributed sparse annotations and explicitly reduce the epistemic uncertainty. Without any prior restrictions and additional information, our proposed method achieves comprehensive performance improvements at multiple label rates and different annotation distributions.