Self-supervised-Fewshot-Medical-Image-Segmentation
Self-supervised-Fewshot-Medical-Image-Segmentation copied to clipboard
cheng-01037
How to set up to train only manual annotation data
Only SSL training is given in the code. How to set it to only train manual annotation data？
Hi did you successfully train on the manual annotation data? Besides, may I ask how do you deal with transforms=None on the validation.py dataset, as on ManualAnnoDataset I notice it only has "img, lb = self.transforms()" which might get error if transforms is None.
Only SSL training is given in the code. How to set it to only train manual annotation data？
Have you experienced this problem when dealing with CT datasets? KID_R and KID_l are empty on my Classmap.json
Thanks for the question and sorry for the late response. Will update in a future version.
Only SSL training is given in the code. How to set it to only train manual annotation data？
Could you please elaborate on this ?
Hi yes will update the code to support manual annotation training but I am afraid that I may not have time to clean up the code in recent days ... sorry about that.
The core idea is still to alleviate over-fitting and to make sure prototype alignment loss not to explode.
Hi yes will update the code to support manual annotation training but I am afraid that I may not have time to clean up the code in recent days ... sorry about that.
The core idea is still to alleviate over-fitting and to make sure prototype alignment loss not to explode.
Hi, I also what to ask how to set it to only train manual annotation data？Is the code avalible?
