Enhancing 3D Semantic Scene Completion with a Refinement Module

TL;DR

This paper introduces ESSC-RM, a versatile refinement framework that enhances 3D semantic scene completion models by improving their prediction accuracy through a two-phase process involving coarse prediction and subsequent refinement.

Contribution

The paper presents a novel plug-and-play refinement module, ESSC-RM, that significantly improves existing SSC models' performance by integrating a multiscale refinement process.

Findings

Improved mean IoU on SemanticKITTI dataset for integrated models

Demonstrated general applicability of ESSC-RM across different SSC architectures

Consistent enhancement of semantic prediction accuracy

Abstract

We propose ESSC-RM, a plug-and-play Enhancing framework for Semantic Scene Completion with a Refinement Module, which can be seamlessly integrated into existing SSC models. ESSC-RM operates in two phases: a baseline SSC network first produces a coarse voxel prediction, which is subsequently refined by a 3D U-Net-based Prediction Noise-Aware Module (PNAM) and Voxel-level Local Geometry Module (VLGM) under multiscale supervision. Experiments on SemanticKITTI show that ESSC-RM consistently improves semantic prediction performance. When integrated into CGFormer and MonoScene, the mean IoU increases from 16.87% to 17.27% and from 11.08% to 11.51%, respectively. These results demonstrate that ESSC-RM serves as a general refinement framework applicable to a wide range of SSC models.