search

Luffy03 / Large-Scale-Medical

[CVPR 2024 Extension] 160K volumes (42M slices) datasets, new segmentation datasets, 31M-1.2B pre-trained models, various pre-training recipes, 50+ downstream tasks implementation
Apache License 2.0
100 stars 7 forks source link

About validation in pretraining #16

Closed jsadu826 closed 2 weeks ago

jsadu826 commented 2 weeks ago

Dear author, for ssl pretraining, I notice there is no validation code. How did you determine the training steps needed and how did you avoid overfitting?

Luffy03 commented 2 weeks ago

I did not conduct validation in SSL pre-training. Since we got plenty of data for pre-training, I think overfitting would not be a big problem.