MONAILabel torchsize error when training with 3d slicer
VSPuzzler opened this issue · 11 comments
Hello. I am trying to use the train feature of MONAILabel using a dataset with masks that were created outside 3dslicer. It starts to do some initializing, but then I get a torch size error. Here is the error I am getting:
I am using a 2070 GPU with an i7 8700k CPU. I was able to run and train MONAILabel before on a different dataset, but I am only having an issue now, so I am confused on what the issue is.
Hi, torch size issue typically refers to the channel wise inconsistency, input data formats, or some parameters errors.
You could double check the original model's usage before using it on additional data, check the inputs channels, image dimensions, etc.
Does this mean there is an issue in the data I am using?
Hi @VSPuzzler,
Does this mean there is an issue in the data I am using?
It could be. We can help if you provide more details about the number of labels, label indexes and whether you've modified the pre-transforms for training.
Are you using a pre-trained model? It could also be that you are using a pre-trained model for segmenting a different number of labels. Try removing whatever is in the radiology/model folder
Hello, I think it was due to a model in the folder that I trained before that was causing the issue Thanks for the help.
Thanks for letting us know, @VSPuzzler
Hello, actually that was not the issue. I re-ran it with with the empty models folder, but I am still having the issue.
Hi, I tried reinstalling monailabel, but that didn't fix anything either.
Hi @VSPuzzler,
I suspect the issue comes from the dataset. Can you please provide more details of the task you are trying to run?
See this comment: #1613 (comment)
Hello, I think there is a problem with the dataset where some of the masks do not match the images. I am trying to fix that right now and will let you know if that fixes it.
Hello, that was the issue. Thank you for the help.
Thanks for letting us know, @VSPuzzler