CA-Net icon indicating copy to clipboard operation
CA-Net copied to clipboard

About label size scaling of the transform.py

Open AI-Hans opened this issue 3 years ago • 1 comments

Should the bilinear interpolation(Image.BILINEAR) of label scaling be changed to nearest neighbor interpolation(Image.NEAREST)?

https://github.com/HiLab-git/CA-Net/blob/94f2624ee6344a960655183b9c1cc8dfb088498a/utils/transform.py#L163

AI-Hans avatar Jan 17 '22 08:01 AI-Hans

Thanks for your issue. Yes, it is better to use nearest-neighbor interpolation for label resize.

JoeGue avatar Feb 11 '22 10:02 JoeGue