View-aligned pixel-level feature aggregation for 3D shape classification

Yong Xu, Shaohui Pan, Ruotao Xu, Haibin Ling

Abstract: Highlights•A novel “fuse-while-extract” pipeline for multi-view 3D model recognition.•A module fuses pixel-level features based on their receptive fields.•A novel method performs feature fusion on both view-level and pixel-level features.•The proposed method achieves SOTA on 3D model classification and retrieval.