DCS-Net: Pioneering Leakage-Free Point Cloud Pretraining Framework with Global Insights

Masked autoencoding and generative pretraining have achieved remarkable success in computer vision and natural language processing, and more recently, they have been extended to the point cloud domain. Nevertheless, existing point cloud models suffer from the issue of information leakage due to the pre-sampling of center points, which leads to trivial proxy tasks for the models. These approaches primarily focus on local feature reconstruction, limiting their ability to capture global patterns within point clouds. In this paper, we argue that …

and natural language processing center cloud computer computer vision cs.cv domain framework free generative global information insights issue language language processing leads natural natural language natural language processing pretraining processing sampling success vision