How do you tackle with  train-test discrepancy?

Congrats and appreciate your great work!

May I ask when the full source code will be updated? And when dealing with segmentation task, how do you mitigate train-test discrepancy/ bias exposure on label and image?

Thanks a lot for your attention.