Channel Pruning for Efficient Convolution Neural Networks