pluo911

Results 5 comments of pluo911

Nan error caused by “N X C X 1 X 1” input features

In fc layer, IN and LN should be the same. R50v2+SN converges much faster than R50v1+SN and produces better top-5 acc.

when I use SN instead of BN, there is a big difference between val acc and train acc

@GYxiaOH Try batch average when evaluating BN in SN. Batch average is stable than moving average for BN. In some tasks there could be difference, please see figure 8 in...

I complete the SN by Keras. welcome to advice

Thanks for your interest. SN benefits from adding 0.5 dropout in the last layer of hidden features, but GN and BN might not. The improvement depends on the generation error...

why not add gn

@Latou GN can be included in SN. You may try GN in your problem.

Syncronized SN

@eugenelawrence We are planning to do this. Welcome to contribute.