Exploring Easy Boosts for Lidar Semantic Scene Completion

Forum topic · 小凯 · 2026-06-04

Summary

This paper investigates "free lunch" strategies to improve lidar semantic scene completion (SSC) without complex architectural redesigns. The authors show that equipping input point clouds with semantic pseudo-labels from off-the-shelf segmentation models significantly boosts existing SSC architectures. Comparison against an oracle confirms that high-quality semantic priors are the primary driver of mIoU gains. Additionally, adding visibility information that distinguishes empty space from unknown space to the input lidar scan provides a secondary performance boost across tested architectures. With these simple enhancements, older SSC models remain competitive with state-of-the-art systems and can even surpass them. The work is by Tetiana Martyniuk, Jonathan Seele, Alexandre Boulch, Gilles Puy, Renaud Marlet, and Raoul de Charette, available on arXiv as 2606.03992.

Overview

Field: Computer Vision
Authors: Tetiana Martyniuk, Jonathan Seele, Alexandre Boulch, Gilles Puy, Renaud Marlet, Raoul de Charette
Published: 2026-06-02
arXiv: 2606.03992

Abstract

This paper investigates "free lunch" strategies to boost the performance of lidar semantic scene completion (SSC) without requiring complex architectural redesigns.

Key contributions:

Semantic pseudo-labels: Endowing input point clouds with semantic pseudo-labels from off-the-shelf segmentors significantly improves the performance of existing SSC architectures.
Oracle analysis: By evaluating these models against an oracle, the authors establish that high-quality semantic priors are a primary driver of mIoU gains.
Visibility information: Equipping the input lidar scan with visibility information that distinguishes between empty and unknown spaces provides a secondary performance boost across the tested architectures.

Findings

Using these simple enhancements, the authors observe that older models remain competitive with state-of-the-art systems, and can even outperform them.

---

*Auto-collected on 2026-06-04.*