Figure 2
A diagram showing the flow of a proposed three D Voxel Backbone Layer and a Refinement stage layer.

An overview of our model structure. The input point cloud will first voxelized and go through three of ours 3D backbone layer to produce RoIs. Then, the convolution features and the predicted foreground score will be used in further refinement stage. In refinement stage, we will utilize the score and calculate the distance to the corresponding grid points as the sample criteria.

or Create an Account

Close subscription notice
Close access options