Interpretable point cloud classification using multiple instance learning
Keywords: Multiple instance learning, point cloud, interpretable
TL;DR: We propose an interpretable point cloud classification framework using multiple instance learning
Abstract: 3D image analysis is crucial in fields such as autonomous driving and biomedical research. However, existing 3D point cloud classification models lack interpretability, limiting trust and usability in safety-critical applications. To address this, we propose PointMIL, an inherently locally interpretable point cloud classifier using Multiple Instance Learning (MIL). PointMIL offers local interpretability, providing fine-grained point-specific explanations to point-based models without the need for \textit{post-hoc} methods, addressing the limitations of global or imprecise interpretability approaches. We applied PointMIL to four popular point cloud classifiers, PointNet, DGCNN, CurveNet, PointMLP, and PointNeXt, and proposed a transformer-based backbone to extract high-quality point-specific features. PointMIL made these models inherently interpretable while increasing predictive performance on standard benchmarks (ModelNet40, ShapeNetPart) and achieving state-of-the-art mACC ($97.3\%$) and F1 ($97.5\%$) on the IntrA biomedical data set, and another dataset of biological cells. To our knowledge, this is the first work to apply MIL to interpretable point cloud classification.
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.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 12357
Loading