The prediction will skip some frame in nnunetv2 3d_fullres mode

[derekshiii]

Thanks for your great work, but i find a problem during the inferece stage. When i used the pretrained model to predict cine MRI, I find the prediction is update 3 or 4 frames in 3d mode while the 2d mode don't have this problem. Hope your reply!

[derekshiii]

I used MMs Dataset to train the nnunet model, this problem only appear in my privite dataset. It won't appear in the test dataset in MMS Dataset. I try modifying the sliding window step but do not works. Here is the GIF which show my problem.

[mrokuss]

Could you check if the file you pass to nnUNet during the prediction has the same resolution as the image you are viewing? This looks to me like the image used by nnUNet to generate a prediction is of worse resolution (in the z-axis especially) and therefore the steps you observe are generated. Hope this helps.

[derekshiii]

Thanks for the help by the contributer! This problem have been soloved, The reason of this problem is the Z axis resolution of my privite dataset is smaller than the MMS dataset