Multi-scale Recursive and Perception-Distortion Controllable Image Super-Resolution

Pablo Navarrete Michelini, Dan Zhu, Hanwen Liu

Abstract: We describe our solution for the PIRM Super–Resolution Challenge 2018 where we achieved the $$\varvec{2^{nd}}$$ best perceptual quality for average $$RMSE\leqslant 16$$ , $$5^{th}$$ best for $$RMSE\leqslant 12.5$$ , and $$7^{th}$$ best for $$RMSE\leqslant 11.5$$ . We modify a recently proposed Multi–Grid Back–Projection (MGBP) architecture to work as a generative system with an input parameter that can control the amount of artificial details in the output. We propose a discriminator for adversarial training with the following novel properties: it is multi–scale that resembles a progressive–GAN; it is recursive that balances the architecture of the generator; and it includes a new layer to capture significant statistics of natural images. Finally, we propose a training strategy that avoids conflicts between reconstruction and perceptual losses. Our configuration uses only 281 k parameters and upscales each image of the competition in 0.2 s in average.