this repository contains the tensorflow-1.x implementation of the recent Cross-Iteration BatchNormalization method. Please visit Cross-iterationBatchNorm if you're interested about the original PyTorch Version. Or you can read arxiv for more details.
I was wondering that why this implementation always has a lower performance compared to the original BN. Is it because the dataset I use is too small or what I did wrong? If you know this please please tell me.
A boring master student from CQUPT. email: hlldmail@qq.com