In this work, we present non-local operator combination with the GCN network for mnist number classification. The mnist dataset will first turn into graph by the top-75 light pixels.We want to test the difference and the performance of different non-local operator comparing to the classifcal GCN network.
-
install pytorch-geometric (You may need to use conda)
-
The training code is inside train.py , you can change the network architecture if you want