Paper 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. The authors 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, they establish that high-quality semantic priors are a primary driver of mIoU gains. Furthermore, they 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, the authors observe that older models remain competitive with state-of-the-art systems, and can even outperform them.
Key Findings
- Semantic pseudo-labels matter: Adding semantic pseudo-labels from off-the-shelf segmenters to input point clouds substantially improves existing SSC architectures.
- Oracle analysis: Comparison with an oracle confirms that high-quality semantic priors are the main source of mIoU improvements.
- Visibility cues help: Annotating input scans with visibility information separating empty from unknown space yields an additional, secondary performance boost.
- Implication: Older SSC models, with these simple input enhancements, remain competitive with — and can even exceed — state-of-the-art systems.
*Auto-collected on 2026-06-04*