publications

2024

1. WACV 2025

   

   Continual Learning in 3D Point Clouds: Employing Spectral Techniques for Exemplar Selection

   Hossein Resani, Behrooz Nasihatkon, and Mohammadreza Alimoradi Jazi

   Accepted at IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024

   Abs PDF Video

   A novel framework for Continual Learning in 3D object classification (CL3D) based on the selection of prototypes from each class using spectral clustering is introduced, and the effectiveness of clustering in the input space, local feature space, and global feature space is explored.

2. ICLR 2025

   

   MIRACLE 3D: Memory-efficient Integrated Robust Approach for Continual Learning on Point Clouds via Shape Model construction

   Hossein Resani, and Behrooz Nasihatkon

   Accepted at International Conference on Learning Representations (ICLR), 2024

   Abs PDF Video

   This paper presents a novel framework for memory-efficient and privacy-preserving continual learning in 3D point cloud classification. Instead of storing numerous exemplars, our method uses a compact shape model that stores only the mean shape and key modes of variation, significantly reducing memory usage while enhancing privacy. The approach includes Gradient Mode Regularization to improve model robustness and stability against input variations.