Enhancing trust in automated 3D point cloud data interpretation through explainable counterfactuals

Andreas Holzinger, Niko Lukac, Dzemail Rozajac, Emile Johnston, Veljka Kocic, Bernhard Hoerl, Christoph Gollob, Arne Nothdurft, Karl Stampfer, Stefan Schweng, Javier Del Ser

Abstract: Highlights•A novel framework balances similarity, sparsity, and validity in 3D counterfactuals.•Demonstrates robustness on real-world noisy LiDAR data for enhanced model trust.•Advances multi-objective optimization for explainable AI in point cloud segmentation.•Introduces efficient counterfactuals to increase transparency in 3D applications.