How to change the way the graph is built in the code

[ghost]

Thank you for your excellent work. It is mentioned in the paper that Knn is used to build the graph. When I read the code, I don't clearly understand which piece of code is used to build the graph. I don't know how this implementation is implemented.

[gyshgx868]

Sorry for my late reply.

The code to construct a kNN graph is in this line:

graph-ter/graph_ter_cls/tools/utils.py

Line 53 in 3b9bda5

indices = knn(x, k=k) # (batch_size, num_points, k)

Simply put, it is firstly to calculate the pairwise distance of all nodes, and then for each node, take the first k nearest neighbor nodes as its neighbors.

This code is directly borrowed from: https://github.com/WangYueFt/dgcnn/tree/master/pytorch