XSPLAIN: XAI-enabling Splat-based Prototype Learning for Attribute-aware INterpretability

Researchers have unveiled XSPLAIN, the first interpretability framework designed specifically for 3D Gaussian Splatting (3DGS) models. It uses a novel prototype-based method to explain AI decisions by showing similar training examples, moving beyond confusing saliency maps. In a user study with 51 participants, XSPLAIN explanations were chosen as the best 48.4% of the time, significantly outperforming all other methods (p<0.001) without sacrificing any classification accuracy.