search

HiLab-git / SSL4MIS

Semi Supervised Learning for Medical Image Segmentation, a collection of literature reviews and code implementations.
MIT License
2.13k stars 381 forks source link

model performance #53

Open wwjwy opened 2 years ago

wwjwy commented 2 years ago

Hi,I have a confusion hope can be helped， In semi-supervised segmentation, why the performance improvement always becomes smaller as the amount of labeled data increases? ~YDFZQ7)L`~JM2M_K T~V`C Thanks!

upupming commented 2 years ago

It is easy to understand, the amount of labeled data increases, the model already achieves good results using the labeled data, and the unlabeled data is less, so the improvement will be much lower.