Fusing batch normalisation and convolution for faster inference

Fusing adjacent convolution (Conv)and batch normalisation (BN)layers is a practical way of boosting inference speed. Batch normalisation is one of the most important regularisation techniques in the modern deep learning field.