Weakly supervised point cloud semantic segmentation has attracted a lot of attention due to its ability to alleviate the heavy reliance on fine-grained annotations of point clouds. However, in practice, sparse annotation usually exhibits a distinct non-uniform distribution in point cloud, which poses challenges for weak supervision. To address these issues, we propose an adaptive annotation distribution method for weakly supervised point cloud semantic segmentation. Specifically, we introduce the probability density function into the gradient sampling approximation analysis and investigate the impact of sparse annotations distributions. Based on our analysis, 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 with different annotation distributions on S3DIS, ScanNetV2 and SemanticKITTI.

翻译：弱监督点云语义分割因能够缓解对点云细粒度标注的严重依赖而受到广泛关注。然而，在实际应用中，稀疏标注通常在点云中呈现明显的非均匀分布，这给弱监督带来了挑战。为解决这些问题，我们提出了一种用于弱监督点云语义分割的自适应标注分布方法。具体而言，我们将概率密度函数引入梯度采样近似分析，并研究了稀疏标注分布的影响。基于分析，我们提出了一种标签感知的点云下采样策略，以提高训练阶段中标注的参与比例。此外，我们设计了乘法动态熵作为梯度校准函数，以缓解由非均匀分布的稀疏标注引起的梯度偏差，并显式降低认知不确定性。无需任何先验限制和额外信息，我们的方法在S3DIS、ScanNetV2和SemanticKITTI数据集上，以不同标注分布的多种标签率实现了全面的性能提升。