Filter Response Normalization Layer: Eliminating Batch Dependence in the Training of Deep Neural Networks | IEEE Conference Publication | IEEE Xplore