SAMBLE: Learning Shape-Specific Sampling Strategies for Point Cloud Shapes with Sparse Attention Map and Adaptive Bin Partitioning
Keywords: 3D Shapes, Point Cloud Sampling, Sparse Attention Map, Bin Partitioning
TL;DR: A novel point cloud sampling method learns shape-specific sampling strategies that strikes superior balance between sampling local details and maintaining global outline.
Abstract: Point cloud sampling plays a pivotal role in facilitating efficient analysis of large-scale point clouds. Recently, learning-to-sample methods have garnered growing interest from the community, particularly for their ability to be jointly trained with downstream tasks. However, previous learning-based sampling methods either lead to unrecognizable sampling patterns by generating a new point cloud or biased sampled results by focusing excessively on shape details. Moreover, they all fail to take the natural point distribution variations over different shapes into consideration and learn a similar sampling strategy for all point clouds. In this paper, we propose a Sparse Attention Map and Bin-based Learning method (termed SAMBLE) to learn shape-specific sampling strategies for point cloud shapes, striking a superior balance between the overall shape outline and intricate local details for the sampling process. In particular, we first propose sparse attention map by integrating both local and global information. Based on this, multiple point-wise sampling score computation methods are proposed and explored by leveraging heatmaps as a guiding tool. Subsequently, we introduce a binning strategy that partitions points within each point cloud based on these scores. Finally, additional learnable tokens are introduced during the attention computation phase to acquire sampling weights for each bin, thereby enabling the development of shape-specific sampling strategies for an optimized sampling process. Extensive experiments demonstrate that our method adeptly strikes a refined balance between sampling edge points for local details and preserving uniformity in the global shape, leading to superior performance across common point cloud downstream tasks and even in scenarios involving few-point cloud sampling.
Primary Area: learning on graphs and other geometries & topologies
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: 8275
Loading