Local information fusion network for 3D shape classification and retrieval

作者：

Highlights：

• We propose a novel framework (LIFN) that considers the local characteristics, region-wise and channel-wise relationships.

• We propose a novel module (ROM) for feature map reorganization by clustering regions from multiple views into super regions.

• We design a novel module (RAM) for region-wise and channel-wise affinities exploration in the super-region feature maps.

摘要

•We propose a novel framework (LIFN) that considers the local characteristics, region-wise and channel-wise relationships.•We propose a novel module (ROM) for feature map reorganization by clustering regions from multiple views into super regions.•We design a novel module (RAM) for region-wise and channel-wise affinities exploration in the super-region feature maps.

论文关键词：3D shape recognition,Attention mechanism,Local information

论文评审过程：Received 21 November 2021, Revised 21 January 2022, Accepted 13 February 2022, Available online 17 February 2022, Version of Record 21 March 2022.

论文官网地址：https://doi.org/10.1016/j.imavis.2022.104405