FH mask

[baiti01]

Thank you for the great work. Was wondering whether why you use the FH mask as extra supervision? Will the performance be affected significantly if remove it? Do you have some ablation studies? Thank you.

[Yuxin-Du-Lab]

Thanks for your support!
Our purpose is to create a universal segmentation method like SAM.
If we remove the FH masks, the model will perform badly for the unlabeled area in CTs due to the serious partial label problem in volumetric medical image segmentation datasets.
But if you just want it to perform well in a fixed category set, getting rid of the FH mask will improve the performance to some degree. Just depends on your needs.

We conducted some ablation studies during the prototype design stage but they are not systematized so we do not have plan to publish them.