search

ziyangwang007 / CV-SSL-MIS

Exploring CNN and ViT for Semi-Supervised Medical Image Segmentation
Apache License 2.0
102 stars 10 forks source link

Test 3D - train_uncertainty_aware_mean_teacher_3D #16

Open carlotita22 opened 3 months ago

carlotita22 commented 3 months ago

Hello! I would really appreciate your help. Thank you very much for the excellent work. I was trying to test the network mentioned in the title using 3UNet (python train_uncertainty_aware_mean_teacher_3D.py --root_path ../data/BraTS2019 --exp BraTS/exp1 --model unet_3D --max_iterations 1000 --batch_size 4 --base_lr 0.001 --labeled_num 20). I was able to train it with the BraTS2019 data; however, I can't use test_3D.py. I was checking the code, and it seems that the folder from test_3D_util import test_all_case is missing.

carlotita22 commented 3 months ago

I found the file ! https://github.com/HiLab-git/SSL4MIS/blob/master/code/test_3D_util.py (you need do changes but its works)

ziyangwang007 commented 3 months ago

Hi, thanks for your message. I will update this repo soon.

Yes, this project is heavily based on SSL4MIS project, and I borrow lots of code. And at the same time, I would like to highlights some differences I feel you might be interested: (1) more datasets (2) more baseline methods (3) some baseline methods with higher performance, such as 'https://github.com/ziyangwang007/CV-SSL-MIS/blob/main/code/train_cnn_meet_vit_2D.py' (4) self-supervised learning.

carlotita22 commented 3 months ago

Thank you very much for your response! I have a question regarding the data (input to the network). First, you use the brats_proprecessing.py code (location: code/dataloaders/brats_proprecessing.py), which handles NIfTI files. Then you convert it to H5 format (I found this reference code: link). Finally, when training, you use dataloaders.brats2019. I’m not very clear on the data preprocessing workflow. I’m asking because I would love to use these steps for my own 3D data (cardiac MRI).