GAM : Gradient Attention Module of Optimization for Point Clouds Analysis
This work addresses a domain-specific problem in 3D point cloud analysis by improving local feature aggregation for tasks like segmentation and classification, though it is incremental as it builds on existing methods.
The paper tackles the problem of local feature aggregation in point cloud analysis by proposing a Gradient Attention Module (GAM) that uses fine-grained geometric information, achieving a 35X speedup and state-of-the-art performance on the S3DIS dataset with mIoU/OA/mAcc of 74.4%/90.6%/83.2%.
In point cloud analysis tasks, the existing local feature aggregation descriptors (LFAD) are unable to fully utilize information in the neighborhood of central points. Previous methods rely solely on Euclidean distance to constrain the local aggregation process, which can be easily affected by abnormal points and cannot adequately fit with the original geometry of the point cloud. We believe that fine-grained geometric information (FGGI) is significant for the aggregation of local features. Therefore, we propose a gradient-based local attention module, termed as Gradient Attention Module (GAM), to address the aforementioned problem. Our proposed GAM simplifies the process that extracts gradient information in the neighborhood and uses the Zenith Angle matrix and Azimuth Angle matrix as explicit representation, which accelerates the module by 35X. Comprehensive experiments were conducted on five benchmark datasets to demonstrate the effectiveness and generalization capability of the proposed GAM for 3D point cloud analysis. Especially on S3DIS dataset, GAM achieves the best performance among current point-based models with mIoU/OA/mAcc of 74.4%/90.6%/83.2%, respectively.