Semantic Occupancy Prediction with Dual Range-Voxel Representation

2026-06-30

AI Summary

This work addresses the challenges of insufficient semantic information and limited geometric awareness in single-frame sparse point clouds for 3D semantic occupancy prediction. To achieve both efficiency and robustness without relying on multi-frame fusion, the authors propose a dual-domain joint modeling approach. It employs a range-view encoder to capture contextual semantics and a geometry-aware multi-scale voxel-view encoder to extract spatial structures, coupled with a bidirectional fusion module that effectively integrates these complementary representations. Evaluated on nuScenes-Occupancy, the method outperforms multi-frame approaches by 5.4% in mIoU while achieving a 2.1× speedup in inference. It also demonstrates state-of-the-art performance on SemanticKITTI and SemanticPOSS benchmarks.

Abstract

LiDAR-based 3D semantic occupancy prediction, which aims to provide accurate and comprehensive scene representation, is crucial for autonomous driving systems. As point clouds suffer from sparsity and incompleteness, leading to insufficient semantic learning and difficult occupancy perception, existing methods often stack multi-sweep point clouds to obtain dense spatial information. However, such a naive strategy also results in efficiency (e.g., additional computational burden) and robustness (e.g., pose transformation noise) concerns, which hinder their practical applications. In this work, we propose a Dual Range-Voxel Representation (DRVR) that leverages the range-view context and voxel-view geometry of single-sweep point clouds for 3D semantic occupancy prediction, eliminating the concerns associated with the multi-sweeps. Specifically, we use the range-view encoder to extract the compact context of the scene. To fully exploit the spatial information, we design a geometry-aware voxel-view encoder that extracts multi-scale voxel-view features separately and combines them for better geometric occupancy prediction. Moreover, we propose a range-voxel fusion module to cooperate range- and voxel-view features via voxel-to-range and range-to-voxel fusions. Extensive experiments on nuScenes-Occupancy, SemanticKITTI and SemanticPOSS show the superiority of our method. Especially on nuScenes-Occupancy, our single-sweep DRVR achieves 5.4% improvement in mIoU and 2.1x acceleration compared to the multi-sweep method.

Problem

• Semantic Occupancy Prediction
• LiDAR Point Clouds
• Sparsity
• Multi-sweep Fusion
• Autonomous Driving

Innovation

• Semantic Occupancy Prediction
• Dual Range-Voxel Representation
• Single-sweep LiDAR
• Range-Voxel Fusion
• 3D Scene Understanding