The paper presents a highly accurate single-image super-resolution (SR) method using a very deep convolutional network inspired by VGG-net. The authors find that increasing the network depth significantly improves accuracy, and their final model uses 20 weight layers. By cascading small filters multiple times, the network efficiently exploits contextual information over large image regions. To address the issue of slow convergence in very deep networks, they propose a training procedure that involves residual-learning and extremely high learning rates (10^4 times higher than SRCNN). The method outperforms existing methods in both accuracy and visual improvements. The paper also discusses the benefits of a single-model approach for multi-scale SR, demonstrating that a single convolutional network can handle multiple scale factors efficiently. Experimental results on several datasets show that the proposed method achieves superior performance compared to state-of-the-art methods.The paper presents a highly accurate single-image super-resolution (SR) method using a very deep convolutional network inspired by VGG-net. The authors find that increasing the network depth significantly improves accuracy, and their final model uses 20 weight layers. By cascading small filters multiple times, the network efficiently exploits contextual information over large image regions. To address the issue of slow convergence in very deep networks, they propose a training procedure that involves residual-learning and extremely high learning rates (10^4 times higher than SRCNN). The method outperforms existing methods in both accuracy and visual improvements. The paper also discusses the benefits of a single-model approach for multi-scale SR, demonstrating that a single convolutional network can handle multiple scale factors efficiently. Experimental results on several datasets show that the proposed method achieves superior performance compared to state-of-the-art methods.