search

ServiceNow / geo-bench-experiments

Apache License 2.0
3 stars 1 forks source link

TypeError when using resnet18 for classification #5

Closed Purin100 closed 3 months ago

Purin100 commented 4 months ago

I generated a classification task following the instruction on the page. Everything was fine at this stage.

When I tried to run the experiment with m-eurosat dataset, I saw the following error message: Problem The pytorch version of my computer is torch 2.3.0 with CUDA 11.8, python version is 3.10.14.

I tried convnext model on the same dataset, this error didn't show up.

It would be appreciated that you can fix this problem. Thank you!

By the way, I find another error when using resnet50 model. It seems that pytorch no longer accepting monument which exists in the resnet50.yaml as a parameter for Adam optimizer. Keeping the parameter will cause an error when running the experiment. I modified optimizer part in resnet50.yaml (on my computer) according to that in resnet18.yaml, and this error got fixed.

Could you please update resnet50.yaml in your project to fix this error? Thank you!

nilsleh commented 3 months ago

Hi @Purin100 , thanks for the issue. The error you are seeing occured, because the torchgeo dependency changed the return statement of the function we used. I just pushed to main, and should be fixed now. I can run classification experiments.

Purin100 commented 3 months ago

Hello @nilsleh, I tested the code on my computer and it worked well. Thank you!