search

uni-medical / STU-Net

The largest pre-trained medical image segmentation model (1.4B parameters) based on the largest public dataset (>100k annotations), up until April 2023.
Apache License 2.0
264 stars 24 forks source link

How to train from scratch(re-pretraining)? #28

Closed eclipse0922 closed 6 months ago

eclipse0922 commented 6 months ago

Thank you for sharing awesome work!

I did fine tuning for my dataset with pre-trained weight(small and base model) It shows me great result.

Can you share the training script used to train the TotalsSegmentator dataset? Or is it sufficient to use the method in this link? https://github.com/wasserth/TotalSegmentator/blob/master/resources/train_nnunet.md

I would like to add more datasets to the pre-training model to maximise performance.

like Verse and CTPelvic1K or more, Of course, I may have to change the information on their labels.

eclipse0922 commented 6 months ago

same question. https://github.com/uni-medical/STU-Net/issues/11