search

jiwoon-ahn / psa

Learning Pixel-level Semantic Affinity with Image-level Supervision for Weakly Supervised Semantic Segmentation, CVPR 2018
MIT License
380 stars 62 forks source link

learning rate #8

Closed LeiyuanMa closed 5 years ago

LeiyuanMa commented 5 years ago

in train_cls.py,when the learning rate is set to 0.1,the loss is increasing,so I tried to set it to 0.01,it works,,so can you verify the learning rate in train_cls.py and train_aff.py?Thanks in advance!

jiwoon-ahn commented 5 years ago

That should work fine. Perhaps there is a problem while loading the pre-trained model. Could you check (https://github.com/jiwoon-ahn/psa/issues/1)?

LeiyuanMa commented 5 years ago

modelA and modelA1 doesn't make difference,for me the learning rate need to be set to 0.01,after train for a night,the loss of validation decreased to 0.079,it seems works. By the way, in train_cls.py,line133 and line 142,the "if...else..."really confused me,I think I understand what you mean,in order to be more clear,could you revise it ?

jiwoon-ahn commented 5 years ago

The learing rate given in the tutorial is only for VGG-16. It is not "if ... else" sentence, rather it is "for ... else" sentence. I don't see any problem in there. But the code is not perfect. It does contain many typos, grammar errors, etc.The best way to fix it is to fork yours.

LeiyuanMa commented 5 years ago

Thanks for your reply,it works fine now.