This paper presents Enhanced Deep Residual Networks (EDSR) and a Multi-Scale Deep Super-Resolution (MDSR) system for single image super-resolution (SISR). The EDSR model improves upon existing state-of-the-art methods by optimizing the residual network architecture by removing unnecessary modules, leading to better performance and computational efficiency. The MDSR system allows for the reconstruction of high-resolution images at different upscaling factors using a single model, reducing model size and training time. Both models are evaluated on benchmark datasets and the NTIRE2017 Super-Resolution Challenge, where they achieve state-of-the-art results. The EDSR+ and MDSR+ versions of the models, which incorporate geometric self-ensemble techniques, further enhance performance. The proposed methods demonstrate superior performance in terms of PSNR and SSIM compared to existing methods, and the MDSR model shows comparable performance to single-scale models while being more efficient. The results show that the proposed models outperform existing methods in both standard benchmark datasets and the NTIRE2017 Super-Resolution Challenge.This paper presents Enhanced Deep Residual Networks (EDSR) and a Multi-Scale Deep Super-Resolution (MDSR) system for single image super-resolution (SISR). The EDSR model improves upon existing state-of-the-art methods by optimizing the residual network architecture by removing unnecessary modules, leading to better performance and computational efficiency. The MDSR system allows for the reconstruction of high-resolution images at different upscaling factors using a single model, reducing model size and training time. Both models are evaluated on benchmark datasets and the NTIRE2017 Super-Resolution Challenge, where they achieve state-of-the-art results. The EDSR+ and MDSR+ versions of the models, which incorporate geometric self-ensemble techniques, further enhance performance. The proposed methods demonstrate superior performance in terms of PSNR and SSIM compared to existing methods, and the MDSR model shows comparable performance to single-scale models while being more efficient. The results show that the proposed models outperform existing methods in both standard benchmark datasets and the NTIRE2017 Super-Resolution Challenge.