Consulting about the loss function

[abcxubu]

Thanks so much for sharing this great project. When I did the experiment about the multi-task-based Semi-supervised learning algorithm, I found the inter-model consistency loss and the inter-task consistency loss were not very helpful for improving algorithm performance. Is this because I use my own dataset (blood vessels)? Could you give me some suggestions? I really appreciate any help you can provide. Hope to hear from you as soon as possible.