search

keras-team / keras-applications

Reference implementations of popular deep learning models.
Other
2k stars 907 forks source link

Missing Relu in DenseNet #31

Closed daavoo closed 6 years ago

daavoo commented 6 years ago

In the official DenseNet implementations (caffe, torch, pytorch) and others (mxnet, torchvision) a Relu activation is applied after the last batch norm and before the global pooling, however this activation does not exist in this keras implementation.

taehoonlee commented 6 years ago

You are right. Thank you for the report, @daavoo. Please see the commit. The top-5 accuracies have been improved by about 0.3% because of your help :)

daavoo commented 6 years ago

I'm glad to hear that! Closing this