RuntimeError: cuDNN error: CUDNN_STATUS_NOT_INITIALIZED

[zxt-triumph]

RuntimeError: cuDNN error: CUDNN_STATUS_NOT_INITIALIZED
I always encounter this problem when I reproduce the experiment.
How can i solve this problem？
Looking forward to your reply.

[zxt-triumph]

By the way, what is the format of the dataset catalog?

[zengmmm00]

I also have this problem. It occurs because loss = criterion(pred, train_annotations) , where pred and train_annotations have different shapes. You can check the values of pred and train_annotations.

[HeartArmy]

@zengmmm00 @zxt-triumph Did you guys manage to solve the issue? If so, do you remember what you did to solve the issue? I am sure I arranged the dataset correctly and am not having success debugging the line loss = criterion(pred, train_annotations). Not sure if it's a CUDA/Pytorch dependency error? Thank you.

[bai101315]

Hello, can you tell me the structure of the OCTA dataset? After downloading the dataset, I couldn't match the dataset structure in the code. I don't know how to handle the OCTA500 dataset to make the code run smoothly. Please give me some suggestions, thank you very much.

[HeartArmy]

I never did get to run IPN-V2 correctly, so maybe my dataset folder structure was incorrect. Or maybe the problem is with the code itself. I never did figure it out and stopped working with this model.

[bai101315]

Thank you for your reply. I have seen other people's methods for processing the OCTA500 dataset. I'm also giving up on this model. Good luck for you!

[HeartArmy]

Wishing you the best too. I would suggest trying to pick recent model that has a more active community/authors or where the github issues is active.