Reflectivity Is All You Need!: Advancing LiDAR Semantic Segmentation

TL;DR

The potential benefits of using calibrated intensity for semantic segmentation in urban environments (SemanticKITTI) and for cross-sensor domain adaptation are demonstrated and the Segment Anything Model (SAM) is tested using reflectivity as input, resulting in improved segmentation masks for LiDAR images.

Abstract

LiDAR semantic segmentation frameworks predominantly use geometry-based features to differentiate objects within a scan. Although these methods excel in scenarios with clear boundaries and distinct shapes, their performance declines in environments where boundaries are indistinct, particularly in off-road contexts. To address this issue, recent advances in 3D segmentation algorithms have aimed to leverage raw LiDAR intensity readings to improve prediction precision. However, despite these advances, existing learning-based models face challenges in linking the complex interactions between raw intensity and variables such as distance, incidence angle, material reflectivity, and atmospheric conditions. Building upon our previous work, this paper explores the advantages of employing calibrated intensity (also referred to as reflectivity) within learning-based LiDAR semantic segmentation frameworks. We start by demonstrating that adding reflectivity as input enhances the LiDAR semantic segmentation model by providing a better data representation. Extensive experimentation with the Rellis-3d off-road dataset shows that replacing intensity with reflectivity results in a 4\% improvement in mean Intersection over Union (mIoU) for off-road scenarios. We demonstrate the potential benefits of using calibrated intensity for semantic segmentation in urban environments (SemanticKITTI) and for cross-sensor domain adaptation. Additionally, we tested the Segment Anything Model (SAM) using reflectivity as input, resulting in improved segmentation masks for LiDAR images.

Figures (6)

• Figure 1: Spherical projection of a) Raw intensity data b) Calibrated intensity for Range and angle of incidence($\alpha$) c) Calibrated intensity for near-range ($\eta$), range and angle of incidence($\alpha$).
• Figure 2: Comparison of raw intensity versus range for the grass class. The figure demonstrates how intensity varies with the range and angle of incidence. It highlights the relationship between $\alpha$ and intensity, indicating that the greatest intensity at each range is achieved when $\alpha$ nears 0, while the lowest intensity occurs when $\alpha$ is approximately $\pi/2$.
• Figure 3: Estimated $\eta(R)$ function for Ouster-64.
• Figure 4: Learning Reflectivity model (Picture modified from SalsaNext10.1007/978-3-030-64559-5_16)
• Figure 5: Sample qualitative results from a test set of Rellis-3D. The image shows a spherical projection of point cloud labels for ground truth, Salsanext-$rxyzi$ and Salsanext-$rxyzn$ predictions, along with labeled point cloud projection and camera projections.