Foveated convolutions: improving spatial transformer networks by modelling the retina

The foveated convolution presents an inductive bias, encouraging the subject of interest to be centred in the output of the attention mechanism, giving significantly improved performance.