test problem

[wwwtttjjj]

nice work！
I obtained good results using a private dataset for a four-class classification [0,100,200,255]. However, I encountered a puzzling issue. When I used the command "pymic_nll test config/unet_trinet.cfg" to run the test set, the generated pred_mask values are not only [0,100,200,255], but essentially include all values from 0 to 255.
I am certain that I used four-class classification during my training and I can see the Dice coefficient for each class.

[taigw]

@wwwtttjjj Hi, I do not know what happened exactly with your case. Have you debugged to find some reasons?

[wwwtttjjj]

yean，I have solved this problem, but I don't know why it happened.
If I input a three-channel jpg image, the test output is the wrong mask.But when I input a single-channel png image, the result of test is my four-category mask