Towards Faster Training of Global Covariance Pooling Networks by Iterative Matrix Square Root Normalization | Read Paper on Bytez