bit-da / bcdm Goto Github PK

Hello, thanks for your brilliant work.

I have a question about your Formula (3) in your paper.

Does the Formula (3) corresponding the function discrepancy_calc() ?

Why you calculate loss_cdd by the function discrepancy_calc() in the file train_bcdm.py, while calculate loss_dis by the function cdd() in the file train_visda.py ?

I only achieved 78.2% accuracy in VisDA2017 data set using your code and parameters, may I ask why?

您好！在研读您的论文代码时，我自己写出来的关于office-31数据集的运行效果不太理想，请问下您可以贴上关于office-31数据集在BCDM这篇论文上的代码文件吗，麻烦您了。

In segmentation task, you uses deeplabv2 as your architecture, but I don't see CRF.