Closed roywei closed 4 years ago
As per user feedback, Keras-MXNet was not able to freeze the running mean and var weights during fine tune.
This PR fix the bug and added unit test Reference: https://discuss.mxnet.io/t/how-freeze-batchnorm-layer-in-symbolblock/3949
Thanks.
sandeep-krishnamurthy
commented
4 years ago sandeep-krishnamurthy
commented
4 years ago
As per user feedback, Keras-MXNet was not able to freeze the running mean and var weights during fine tune.
This PR fix the bug and added unit test Reference: https://discuss.mxnet.io/t/how-freeze-batchnorm-layer-in-symbolblock/3949