Kaiseem / DAR-UNet

[JBHI2022] A novel 3D unsupervised domain adaptation framework for cross-modality medical image segmentation
Apache License 2.0
40 stars 5 forks source link

Normalization Range [0,1] or [-1,1] #4

Closed YoungYan-0 closed 1 year ago

YoungYan-0 commented 1 year ago

Dear Dr. Yao,

I noticed that in the paper it says that "We perform min-max normalization to rescale the images intensity to the the range [−1,1]".

However, in the README.md inside the datasets folder, it says that "Normalize each 3D images to [0, 1]".

Which range should I use or does it matter?

Best Regards,

Yucheng

Kaiseem commented 1 year ago

Hi,

Please follow the READMI.md to normalize the 3D images to [0, 1], which is the preparation of the datasets. The preprocessing is in the https://github.com/Kaiseem/DAR-UNet/blob/68a5b994374ba66b0bf72225080a39bc5dbd06f3/utils/dataloader3d.py#L81 (rescale the intensity to [-1, 1]), which is consistent to our paper statement.

If you have any other question, pls let me know. :)

YoungYan-0 commented 1 year ago

Thank you!