Non-stationary deep network for restoration of non-Stationary lens blur.

•This work presents non-stationary deep networks for restoration of non-stationary blur in images due to optical aberrations.•Non-stationary networks utilize location dependent features to cope with the non-stationarity of blur that spans the entire image.•Experimental results show that even shallower non-stationary networks outperform deeper stationary networks.•Networks are trained from pairs of photographed images, eliminating the necessity of estimation of degradation.