Hierarchical Graph Attention Based Multi-View Convolutional Neural Network for 3D Object Recognition

For multi-view convolutional neural network based 3D object recognition, how to fuse the information of multiple views is a key factor affecting recognition performance. Most traditional methods use max-pooling algorithm obtain final feature, which does not take into account correlative between different views. To make full effective views, this paper introduces hierarchical graph attention for recognition. At first, view selection module proposed reduce redundant in can select projective with more information. Then, correlation weighted feature aggregation better features. Finally, structure designed further Extensive experimental results have validated effectiveness method.