Counterfactual Explanations for 3D Point-Cloud Classifiers
Keywords: explainability, counterfactual, 3D, point-cloud
TL;DR: The paper introduces the first method for generating Counterfactual Explanations for 3D point cloud classifiers using a diffusion model and new evaluation metrics.
Abstract: Explainable AI (XAI) seeks to tackle the opacity of deep neural network decisions. Moving beyond the conventional focus on 2D imagery, our research provides the first method to provide Counterfactual Explanations (CEs) for 3D point cloud classifiers. Specifically, we introduce two strategies for 3D CEs using a diffusion model to generate CEs that maintain both semantic consistency and data fidelity in 3D contexts. To this end, we devise novel losses and constraints to boost the realism and practicality of counterfactual instances. Furthermore, we establish a new benchmark with evaluation metrics designed specifically for 3D point clouds allowing future methods to be assessed using it. Altogether, our contributions bridge a key gap in the field of explainability, steering towards more transparent and fair AI methodologies.
Primary Area: interpretability and explainable AI
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 5102
Loading