Closed limzh123 closed 2 years ago
Hey,
thanks for the message and for raising the issue. It's a bit tough to debug based on the above output alone, but I think I have an idea what happened.
I think your pre-trained model that you load with --resume is trained on FashionMNIST, but the default dataset that is loaded in the cmdparser is MNIST. What happens now is that it tries to evaluate and get a "correctly classified tensor" (z embedding), but of course this is empty because it does not really correctly classify anything.
We have checked for this scenario in various functions with "isinstance" checks to see if there are empty lists or tensors present. But it does look like in the visualization line that you pasted in your error, we have forgotten this check.
Could you see if it works if you just re-execute your above command with an additional "--dataset FashionMNIST" (to not only indicate the open set dataset you want to check, but also which dataset has been trained on originally).
If this solves the problem and my above hypothesis is correct, I will include the additional mentioned type check so if the same crash happens again the user can get informed
Hi, It can work after I added the command --dataset FashionMNIST
Thanks a lot and have a nice day!
Great, good to know. Happy I could help.
Hi,
I am encountered this issue when I run this command. !python3 eval_openset.py --resume /content/OpenVAE_ContinualLearning/runs/2022-06-27_12-51-44_FashionMNIST_WRN_variational_samples_1_latent_dim_60/model_best.pth.tar --openset-datasets 'SVHN' --var-samples 100 --calc-reconstruction True --epochs 1
The error is:
Traceback (most recent call last): File "eval_openset.py", line 315, in
main()
File "eval_openset.py", line 112, in main
mean_zs_tensor = torch.stack(mean_zs, dim=0)
TypeError: expected Tensor as element 9 in argument 0, but got list
Thanks if you could assist me