2019 · 97 citations · 19 references
ComputingComputer ScienceComputer VisionMachine LearningMachine VisionDeep LearningScene ReconstructionScene UnderstandingSpatial ResolutionBenchmark DatasetsConvolutional Neural NetworksDataset DescriptionsImage Super-resolutionSuper-resolutionSuper-resolution ImagingSuper-resolution MappingUpscaling
Super-resolution methods in literature has in recent years been dominated by convolutional neural networks (CNNs), aiming to learn a direct mapping from a low to high resolution image. Although successful, these methods rely on large-scale and high-quality datasets, to learn more powerful models. The scale of the existing super-resolution datasets are limiting the performance of current deep and highly complex architectures. Moreover, current datasets are severely limited in terms of resolution, prohibiting the move towards more extreme conditions with high upscaling factors. In this paper, we introduce the DIVerse 8K resolution image dataset (DIV8K). The dataset contains a over 1500 images with a resolution up to 8K. It highly covers diverse scene contents. It is therefore the ideal dataset for training and benchmarking super-resolution approaches, applicable to extreme upscaling factors of 32x and beyond. The dataset was employed for the AIM 2019 Image Extreme Super Resolution Challenge.
19
Image quality assessment: from error visibility to structural similarity
Zhou Wang, Alan C. Bovik, Hamid R. Sheikh et al. · IEEE Transactions on Image Processing · 2004 · 56K citations
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
Christian Ledig, Lucas Theis, Ferenc Huszár et al. · 2017 · 12.3K citations
Image Super-Resolution Via Sparse Representation
John Wright, Thomas S. Huang, Yi Ma · IEEE Transactions on Image Processing · 2010 · 5.3K citations