hfslyc/AdvSemiSeg

关于判别器的输出

Opened this issue · 1 comments

想请问一下在训练得到confidence map过程中判别器的输出map大概是怎样的呢,我现在一直找不到分割网络和判别网路的训练平衡点,我做的肝脏病灶分割,分割网络预先训练使其达到一定的分割效果才加的判别器,判别器loss会降到很低,加上判别器分割精度会下降,请问您对我的训练有什么建议吗

Hi, I'm not experienced with medical datasets so I'm not sure I have good insight for your project. If you can't maintain a good balance between your G and D, you may want to lower the learning rate of D or the lambda for adversarial learning.