Taming the Light: Illumination-Invariant Semantic 3DGS-SLAM

TL;DR

This work tackles illumination sensitivity in 3D Gaussian Splatting SLAM by introducing a proactive Intrinsic Appearance Normalization (IAN) that quantizes albedo to a canonical palette, and a reactive Dynamic Radiance Balancing Loss (DRB-Loss) that adaptively corrects extreme exposure frames. By jointly optimizing geometry, appearance, and semantics with differentiable Gaussian rendering, the approach achieves robust camera tracking, high-fidelity maps, and accurate semantic labeling under varied lighting. Experiments on Replica and ScanNet demonstrate state-of-the-art tracking, rendering quality, and mIoU, highlighting the method's practical impact for real-world robotics and AR. The proposed combination of proactive normalization and reactive correction provides a significant advance toward illumination-invariant 3D semantic mapping.

Abstract

Extreme exposure degrades both the 3D map reconstruction and semantic segmentation accuracy, which is particularly detrimental to tightly-coupled systems. To achieve illumination invariance, we propose a novel semantic SLAM framework with two designs. First, the Intrinsic Appearance Normalization (IAN) module proactively disentangles the scene's intrinsic properties, such as albedo, from transient lighting. By learning a standardized, illumination-invariant appearance model, it assigns a stable and consistent color representation to each Gaussian primitive. Second, the Dynamic Radiance Balancing Loss (DRB-Loss) reactively handles frames with extreme exposure. It activates only when an image's exposure is poor, operating directly on the radiance field to guide targeted optimization. This prevents error accumulation from extreme lighting without compromising performance under normal conditions. The synergy between IAN's proactive invariance and DRB-Loss's reactive correction endows our system with unprecedented robustness. Evaluations on public datasets demonstrate state-of-the-art performance in camera tracking, map quality, and semantic and geometric accuracy.

Figures (3)

• Figure 1: The framework processes input frames by first disentangling albedo and illumination with the IAN module. It then performs camera tracking, using the DRB-Loss to robustly handle exposure variations. Finally, the system jointly optimizes a dense, high-fidelity map containing both geometric and semantic information.
• Figure 2: IAN module's effect on intrinsic appearance.
• Figure 3: Visualization of Our Rendering Performance on the Replica Dataset.