Contextual Range-View Projection for 3D LiDAR Point Clouds

TL;DR

The paper tackles the persistent many-to-one problem in range-view LiDAR projections by introducing context-aware training-time strategies that leverage instance centroids and class labels. Centerness-Aware Projection (CAP) uses a 3D Gaussian centerness score $f(oldsymbol{p}_i) = \frac{1}{(2\pi)^{3/2}} \exp\left(-\frac{1}{2}\|\mathbf{p}_i-\boldsymbol{\mu}\|^2\right)$ with instance center $\boldsymbol{\mu}$ to compute $s_i = \|\mathbf{p}_i\|_2 \cdot \frac{1}{f(\mathbf{p}_i)+\varepsilon}$ and selects the point with the smallest $s_i$ per pixel, thereby preserving central object structure. Class-Weighted-Aware Projection (CWAP) assigns class-specific weights $w_i = w[\text{class}_i]$ and uses $s_i = \|\mathbf{p}_i\|_2 \cdot \frac{1}{w_i+\varepsilon}$ to bias point retention toward target classes, including the option for negative weights. Evaluations on SemanticKITTI with RangeViT show CAP improves instance-class mIoU by up to 3.1%, while CWAP boosts targeted class performance and can trade off others. The methods are training-time enhancements that produce better range-image representations and, consequently, improved downstream semantic segmentation without requiring changes at inference time.

Abstract

Range-view projection provides an efficient method for transforming 3D LiDAR point clouds into 2D range image representations, enabling effective processing with 2D deep learning models. However, a major challenge in this projection is the many-to-one conflict, where multiple 3D points are mapped onto the same pixel in the range image, requiring a selection strategy. Existing approaches typically retain the point with the smallest depth (closest to the LiDAR), disregarding semantic relevance and object structure, which leads to the loss of important contextual information. In this paper, we extend the depth-based selection rule by incorporating contextual information from both instance centers and class labels, introducing two mechanisms: \textit{Centerness-Aware Projection (CAP)} and \textit{Class-Weighted-Aware Projection (CWAP)}. In CAP, point depths are adjusted according to their distance from the instance center, thereby prioritizing central instance points over noisy boundary and background points. In CWAP, object classes are prioritized through user-defined weights, offering flexibility in the projection strategy. Our evaluations on the SemanticKITTI dataset show that CAP preserves more instance points during projection, achieving up to a 3.1\% mIoU improvement compared to the baseline. Furthermore, CWAP enhances the performance of targeted classes while having a negligible impact on the performance of other classes

Figures (4)

• Figure 1: Spherical coordinate representation of a LiDAR measurement, defined by range $r$, azimuth angle $\theta$, and elevation angle $\phi$.
• Figure 2: Overview of the proposed contextual range-view projection pipeline. A 3D LiDAR point cloud is projected into a 2D range image using either Centerness-Aware Projection (CAP) or Class-Weighted-Aware Projection (CWAP), which is then processed by a semantic segmentation network to achieve improved performance.
• Figure 3: Effect of Centerness-Aware Projection (CAP) on range image formation for a single frame. (Top) Standard depth-based projection. (Middle) Result of the proposed CAP. (Bottom) Pixel-wise differences between the two. White pixels indicate unlabeled regions, black pixels correspond to areas with no projected points, and light blue pixels represent cars.
• Figure 4: Effect of CAP and CWAP on range image formation for multiple frames. (Top) Standard depth-based projection. (Middle) Result of the proposed CAP or CWAP. (Bottom) Pixel-wise differences between the two.