Proper way to create and train a low resolution model

[davidkflau]

Hi team,

I have a dataset with image size 512x512x32 and spacing 0.43x0.43x5. I want to train a low-resolution model with image size 128x128x32 and spacing 1.72x1.72x5. We don't have GPU to run inference so low-resolution model will take a faster inference run time. What's the proper way to do this?

When I ran the preprocess and plan, the plans file doesn't include the low-resolution configuration. I found that spacing_increase_factor=1.03 controls the resampling scale.

Should I resample the dataset manually before feeding to the model or change the value of spacing_increase_factor variable (e.g. set it to 4 to resample the image from 0.43x0.43x5 to 1.72x1.72x5)?

Thanks!

[davidkflau]

Hi @dojoh , could you please help with the above question? Thanks!