Answers are grounded in this paper and cite their sources.
Gaussian Splatting algorithms achieve high fidelity under controlled, static conditions with unobstructed viewpoints. However, in unstructured environments hampered by heterogeneous occlusions with sub-optimal lighting conditions or completely dark surroundings, the movement of a co-located camera and light source causes shadow variance as the relative position of objects within the scene produce intricate cast shadows causing the same coordinate of the scene to exhibit different pixel color. This dynamic shifting of shadows leads to transient pixel intensities across the multi-view images, directly violating the photo-consistency assumption inducing the uncertainty for each Gaussian to exhibit varying colors. To address this, we propose SGS: Shadow-Aware Gaussian Splatting, a novel shadow-aware method which explicitly separates intrinsic radiance from extrinsic illumination. Instead of storing separate color sets, each Gaussian is augmented with a single learnable scalar that modulates its intrinsic color features. Furthermore, we introduce a dual-pass rendering pipeline complemented with a bimodal selection loss for an effective disentanglement of shadows from scene geometry. Extensive experiments demonstrate that SGS significantly eliminates volumetric artifacts, while simultaneously increasing overall point coverage complemented with enhanced photometric reconstruction outperforming in all metrics, striking a favorable balance between geometric accuracy, point efficiency and rendering quality.
No takes yet. Share an insight, caveat, or question.
Petrovska et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: