search

hellochick / ICNet-tensorflow

TensorFlow-based implementation of "ICNet for Real-Time Semantic Segmentation on High-Resolution Images".
405 stars 153 forks source link

Apply Dropout to this model #67

Closed alexw92 closed 6 years ago

alexw92 commented 6 years ago

First of all, thank you @hellochick for sharing this great code :+1:

I wonder how to achieve better train results and I noticed that the dropout()- code from network.py is actually never used in model. Didn't you find dropout necessary for this model or is there any other reason (bad results or implementation details) why you didn't use it?

hellochick commented 6 years ago

Hey @alexw92 ,

Sorry for my late reply, thanks for your appreciation. I didn't use the dropout layer since I tried to re-produced the setting in the original paper. But you can try to add dropout layer when you find there exists overfitting problems in your own dataset!

alexw92 commented 6 years ago

Thanks for your response @hellochick ! I agree with you and the authors. Dropout is not necessary in this model and I was able to fix my overfitting issues by just applying validation during training