As we all know, KAN has a very slow training speed and weak task performance. We aim to optimize the training speed of KAN and improve the performance on point cloud tasks.
Due to the surge in popularity of KAN, we plan to incorporate KAN into point cloud-related models for evaluation. We will integrate KAN into familiar point cloud model architectures such as Point-MAE, Point Transformer, and PointNeXt to enhance interpretability and performance. Stay tuned to see what challenges we encounter and how we effectively improve them!