about how to add GCN into my model

[gogowhy]

hello, when I'm implementing my own network(res101)with the GCN part attached , when I start training, the pytorch loss always happen to turn Nans, I watched the implementation of your code, which used mxnet and functions, I want to ask if something needs to be especially prepared when I put the GCN in a ResNet block(class but not function and without mxnet).

[hanwinbi]

I met the same issue with you, but I ran it on 3D-Unet, I came across this whenever I add the global_reasoning_unit to the network, it work fine without it.A year later, the project got no response. Have you solve it?

[BruceYu-Bit]

I also met this problem,how to solve it