patch-based inference approach for 3D volumes

[himashi92]

Hi,

Thanks for sharing your work with the community. Excellent work!

I have successfully trained SAM-Med3D for the brain tumor dataset, and now I want to evaluate the test set. In the paper, it is mentioned that SAM-Med3D operates using a patch-based inference approach. However, in the repository, I needed help finding the script or suitable function you used for 3D patch-based inference. I would appreciate it if you could share the script for this or let me know if I missed something here. :)

Kind regards,
Himashi

[hagersalehahmed]

how can solve this problem File
line 343, in train_epoch
epoch_loss /= step
ZeroDivisionError: float division by zero

[blueyo0]

Hi,

Thanks for sharing your work with the community. Excellent work!

I have successfully trained SAM-Med3D for the brain tumor dataset, and now I want to evaluate the test set. In the paper, it is mentioned that SAM-Med3D operates using a patch-based inference approach. However, in the repository, I needed help finding the script or suitable function you used for 3D patch-based inference. I would appreciate it if you could share the script for this or let me know if I missed something here. :)

Kind regards, Himashi

Hi, Himashi, thank you for your recognition~

Our patch-based inference script is the validation.py, and you can refer to evaluation in Readme for usage. BTW, this script only supports inference on a single patch (crop from the center of the target) instead of sliding-window inference of the whole image. If you need to conduct sliding-window inference, you may need to modify the script yourself or wait for my further update.

[blueyo0]

how can solve this problem File line 343, in train_epoch epoch_loss /= step ZeroDivisionError: float division by zero

Hi, hagersalehahmed
It seems like this issue is raised because your dataset is empty, so no step is run (step=0).
Please check your data first, example data structure in training may be helpful.