Single-Image super-resolution - When model adaptation matters

In recent years, impressive advances have been made in single-image super-resolution. Deep learning is behind much of this success. Deep(er) architecture design and external prior modeling are the key ingredients. The internal contents low-resolution input image neglected with deep modeling, despite earlier works that show power using such priors. paper, we propose a variation residual convolutional neural networks, which has carefully designed for robustness efficiency both testing. Moreover, multiple strategies model adaptation to analyze their strong points weaknesses. By trading runtime priors, achieve improvements from 0.1 0.3 dB PSNR over reported results on standard datasets. Our especially favors images repetitive structures or high resolutions. It indicates more practical usage when our adaption approach applies sequences videos adjacent frames strongly correlated contents. can be combined other simple techniques, as back-projection enhanced prediction, realize further improvements.