Residue Point Cloud Concatenation

Point Cloud is a kind of data which can be used to describe the 3D object, because of the limitaion of the Lidar. Sometimes we can only get some residue point cloud data.

Combine the ideas of MVCNN and DGCNN to make full use of the residue point cloud data.

My network looks like this. By this network, I combained the multi views of the residue point cloud. In the retrieval task, I also pointed out a new loss function to enhance the accuracy. Finally, the accuracy imporved by 4.6%.