RW-Net: Enhancing Few-Shot Point Cloud Classification With a Wavelet Transform Projection-Based Network

In the domain of 3D object classification, a fundamental challenge lies in addressing the scarcity of labeled data, which limits the applicability of traditional data-intensive learning paradigms. This challenge is particularly pronounced in few-shot learning scenarios, where the objective is to achieve robust generalization from minimal annotated samples. To overcome these limitations, it is crucial to identify and leverage the mostsalientanddiscriminativefeatures of 3D objects, thereby enhancing learning efficiency and reducing dependency on large-scale labeled datasets. In this work, we introduce RW-Net, a novel framework designed to address the aforementioned challenges by integrating Rate-Distortion Explanation (RDE) and wavelet transform into a state-of-the-art projection-based 3D object classification architecture. Instead of utilizing RDE as a tool to explain the decision-making process of a given pre-trained classification model as in previous literature, we adapt RDE to train a classification model from scratch to extract critical input features by identifying and preserving the mostinformativedata components while simultaneously reducing redundancy. As RDE operates in wavelet frequency domain, we therefore incorporate wavelet attention to enhance our method's capability to generalize in low-data regimes by emphasizing low-frequency components of the input data and capturing fundamentalgeometricandstructuralattributes of 3D objects. To validate the effectiveness of our RW-Net, we conduct extensive experiments on three benchmark datasets: ModelNet40, ModelNet40-C, and ScanObjectNN for few-shot 3D object classification. The results demonstrate that our approach not only achieves state-of-the-art performance under challenging conditions such as data corruption and cross-domain variations but also exhibits superior generalization and robustness in few-shot learning scenarios. Furthermore, we observe consistent improvements in classification accuracy, even under challenging conditions such as data corruption and cross-domain variations. We release our code to contribute a tool to the research community for advancing few-shot 3D object classification.