Exploring Easy Boosts for Lidar Semantic Scene Completion

2026-06-02

AI Summary

This work addresses the task of LiDAR-based semantic scene completion (SSC) by proposing a simple yet effective input enhancement strategy that requires no architectural modifications to existing models. By augmenting the input point cloud with semantic pseudo-labels and visibility cues generated from off-the-shelf segmentation models, the method effectively distinguishes between empty and unknown space while injecting high-quality semantic priors. This approach seamlessly integrates into prevailing SSC frameworks and consistently yields substantial improvements in mean Intersection over Union (mIoU) across multiple established architectures. Notably, it significantly narrows the performance gap to oracle settings—where ground-truth semantics are assumed available—and enables older models to match or even surpass current state-of-the-art methods, thereby demonstrating the efficacy and broad applicability of this lightweight input enhancement paradigm.

Abstract

This paper investigates "free lunch" strategies to boost the performance of lidar semantic scene completion (SSC) without requiring complex architectural redesigns. We first demonstrate that endowing input point clouds with semantic pseudo-labels from off-the-shelf segmentors significantly improves the performance of existing architectures. By evaluating these models against an oracle, we establish that high-quality semantic priors are a primary driver of mIoU gains. Furthermore, we equip the input lidar scan with visibility information that distinguishes between empty and unknown spaces, which provides a secondary performance boost across the tested architectures. Using these simple enhancements, we observe that older models remain competitive with state-of-the-art systems, and can even outperform them. Our code is available at https://github.com/astra-vision/SSC-Priors.

Problem

• lidar
• semantic scene completion
• performance boost
• semantic priors
• visibility information

Innovation

• semantic scene completion
• lidar
• semantic priors
• visibility information
• pseudo-labels