Single image super-resolution (SISR), as a fundamental low-level vision problem, has been evaluated in the research community and AI industry. Currently, most methods for reconstructing super-resolution images using deep learning are based on models. Training and reconstruction for different types of images is not differentiated at the model level, which may result in waste of resources. At the same time, because of the particularity of the model training of deep learning, the training process is often atomic, which is a great obstacle to the speed of training. We try to build a framework at the system level to further improve the training speed of the model on the basis of ensuring the reconstruction effect. In this paper, a multi-model super-resolution framework is proposed to choosing training network for different input images according to their characteristics, and all input images are classified by a technique named TVAT (Total Variance above the Threshold). Experimental results indicate that our framework brings a 60% performance speedup on average and exhibits good scalability.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.