2016 · 7.6K citations · 22 references
ComputingComputer ScienceComputer VisionMachine LearningMachine VisionDeep LearningImage EnhancementImage QualityRepresentation LearningClippingConvergenceDeep Image FeaturesFiltersImage Super-resolutionResidual NetworksSuper-resolutionSuper-resolution Imaging
Convergence speed becomes a critical issue during training with very deep networks. The authors present a highly accurate single‑image superresolution method and propose a simple yet effective training procedure. The method employs a 20‑layer very deep convolutional network inspired by VGG‑net, learns residuals with extremely high learning rates enabled by adjustable gradient clipping, and cascades small filters to exploit contextual information over large image regions. Increasing network depth significantly improves accuracy, and the proposed method outperforms existing methods with noticeable visual improvements.
We present a highly accurate single-image superresolution (SR) method. Our method uses a very deep convolutional network inspired by VGG-net used for ImageNet classification [19]. We find increasing our network depth shows a significant improvement in accuracy. Our final model uses 20 weight layers. By cascading small filters many times in a deep network structure, contextual information over large image regions is exploited in an efficient way. With very deep networks, however, convergence speed becomes a critical issue during training. We propose a simple yet effective training procedure. We learn residuals only and use extremely high learning rates (104 times higher than SRCNN [6]) enabled by adjustable gradient clipping. Our proposed method performs better than existing methods in accuracy and visual improvements in our results are easily noticeable.
22
Gradient-based learning applied to document recognition
Yann LeCun, Léon Bottou, Yoshua Bengio et al. · Proceedings of the IEEE · 1998 · 58.2K citations · Full text
Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification
Kaiming He, Xiangyu Zhang, Shaoqing Ren et al. · 2015 · 18.8K citations