Collaborate with transforms.Normalize

[TumVink]

Heyy,

Thanks for the great work!

However, I am confused about how people should collaborate with your RandStainNA() and transforms.Normalize() techniques?

Is RandStainNA() an alternative solution for transformations?Normalize()? If so, are we supposed to use this technique as well during the test?

BW,
Jingsong

[TumVink]

@yiqings

[yiqings]

Both shall work as shown in the manuscript.

For the second option (where you use RandStainNA() an alternative solution for transformations.Normalize()), for the test time you can either use 1. raw image 2. RandStainNA() together with Test-time augmentation (where the images is augmented with RandStainNA for multiple times) 3. use the hyper-parameter that decides the mean/std of the distribution for normalization. @TumVink

[yiqings]

Both shall work as shown in the manuscript.

For the second option (where you use RandStainNA() an alternative solution for transformations.Normalize()), for the test time you can either use 1. raw image 2. RandStainNA() together with Test-time augmentation (where the images is augmented with RandStainNA for multiple times) 3. use the hyper-parameter that decides the mean/std of the distribution for normalization. @TumVink

The first option is easy to implement, second can offer some extra performance boost, the third is an trade-off between 1 and 2 based on our experiments.