ROI size sliding window

[kissievendor]

Hi,
I was wondering how UNet deals with the sliding window input.
Because the ROI you set is bigger than the patches UNet is trained on.
How does this work?

Thanks.
Kirsten

[Nic-Ma]

Hi @wyli ,

Could you please help explain something in theory from the researcher aspect?

Thanks.

[YuanTingHsieh]

From what I understand, UNet is consists of CNNs.
So it can take input of different shape.
Like you trained it in [96, 96, 96], but you can inference it in [160,160,160] cubes.

The advantage of inference it using larger cubes is that you get faster inference speed.

Please correct me if I am wrong @wyli

[kissievendor]

Thank you. Do you have some sort of article? Because everything I am reading suggest to use same size. Also, when using a larger sliding window, my results can differ dramatically.