Yunpeng

@terrychenism We refined the batch normalization layers as suggested by [1]. In [1], the authors refine the BN layers by computing the average (not moving average) on a sufficiently large...

@terrychenism =_=!! Nope, It only improved the Top-5 by about 0.03%. Besides, it has some negative effect on Top-1 accuracy. ( Actually, the original Top-1 accuracy is a little bit...

Thank you for your interest. You can use `mx.model.load_checkpoint` to load the complete network definition and use `mx.viz.plot_network` to visualize it. As for the code for defining CRU-Net, I was...

This repo uses an old version of MXNet with a customized [ChannelwiseConvolution](https://github.com/cypw/mxnet/blob/8d6ce66ca47ee3057f5681dcf83b6a329b1dff46/src/operator/channelwise_convolution-inl.h) layer. You may want to use [this MXNet version](https://github.com/cypw/mxnet/tree/8d6ce66ca47ee3057f5681dcf83b6a329b1dff46) to load the pre-trained model.

I guess the mismatched feature map size is caused by the Pooling layer, because Caffe and MXNet use different default rounding strategy for the Pooling layer. So, you may need...