How to use the SDF for mutli-label mask?

JunMa11 / SegWithDistMap

How Distance Transform Maps Boost Segmentation CNNs: An Empirical Study

https://openreview.net/forum?id=hM4pNbXWst

Apache License 2.0

378 stars 67 forks source link

How to use the SDF for mutli-label mask? #15

Closed SkyeC77 closed 3 years ago

SkyeC77 commented 3 years ago

I want to regress the signed distance map with a new head, but it's for a binary mask in your paper. I wonder what should I do if I want to use it for a multi-label mask? is it possible to get the SDF image for a three-label ground truth? like the BraTS dataset. Hope for your reply.

JunMa11 commented 3 years ago

Hi @SkyeC77 ,

I apologize for the very late reply. Recently, I'm busy with the FLARE21 Challenge.

Actually, the SDF implementation should support multi-class segmentation.

https://github.com/JunMa11/SegWithDistMap/blob/153dabf3bc5d9e48058e1497857ac6b00c7abab8/code/train_LA_MultiHead_SDF_L1.py#L76