search

YazhouZhu19 / RPT

[MICCAI 2023] Few-Shot Medical Image Segmentation via a Region-enhanced Prototypical Transformer
34 stars 5 forks source link

Training and testing problems on SABS dataset #7

Closed WangBingJian233 closed 3 months ago

WangBingJian233 commented 5 months ago

Dear author, when I conducted the SABS dataset experiment with your code and the pre-treated SABS hypervoxels you provided, the test results are SPLEEN, RK, LK and GALLBLADDER, but not SPLEEN, RK, LK and Liver as mentioned in this paper. Besides, GALLBLADDER has a very low DICE score on organ segmentation. What's the matter? Is the dataset == 'SABS' part of def get_label_names(dataset) in dataset_specifics.py incorrect (as shown in the figure below)? Could you please modify it? image

yunfeixie233 commented 5 months ago

Hi @WangBingJian233,

I came across your issue here as I'm facing a similar problem. Just wondering, were you able to find a solution?

YazhouZhu19 commented 3 months ago

Hi, Sorry for the delay in replying and thank you for attention. I have updated the files: models/fewshot.py train.py test.py and uploaded the trained models for three datasets under setting 1 and setting 2.

Sorry for not giving the running scripts for SABS datasets, we upload the scripts train_SABS_setting1.sh and train_SABS_Setting2.sh for training model on the SABS dataset. We just employ SPLEEN, RK, LK and Liver these four organs as the preliminary attempt for the few-shot medical image segmentation, and more work is doing. We are grateful that you are interested in our work! Any questions feel free to contact me, and thanks again!