I settled all errors and run this project sucessfully, this is my github forked version

[1suancaiyu]

https://github.com/1suancaiyu/AS-GCN

[yypnogg1011]

hi,Thank you for your work.
But I need to remind you some suggestions，First, the command “cp torchlight/torchlight/init_.py gpu.py io.py ../ ”in your readme file，the codefile name " init" is missing an underscore"_".
Second, I noticed that because the code cannot successfully install the torchlight package like the ST-GCN project, you commented torchlight.gpu.occupy_gpu(gpus) in "processor.io"file to solve this problem，but it maybe cause some people to mistakenly believe that they are using multiple graphics cards for training,I suggest to declare this in Readme file.By the way ,this problem can be solved by manually importing the path.

[YHY-10]

https://github.com/1suancaiyu/AS-GCN

hello，when I was running your project, I came across this problem "pickle.UnpicklingError: pickle data was truncated".Can you tell me how to solve this problem?

[github2016-yuan]

@1suancaiyu Hello, I use your repo and it really works. BUT I got nan error in trainning.Below are steps.

I choose about 16K images from ntu-d (60 classes)

step1:
PretrainAIM: python main.py recognition -c config/as_gcn/ntu-xsub/train_aim.yaml --device 0

It works and the loss is normal.

step2:
TrainMainPipeline: python main.py recognition -c config/as_gcn/ntu-xsub/train.yaml --device 0 --batch_size 4

The loss is always nan.

I try a lot but still can not solve it. So mind you give me some advice? Thx.