Enhanced Deep Residual Networks for Single Image Super-Resolution

10 Jul 2017  ·  Bee Lim, Sanghyun Son, Heewon Kim, Seungjun Nah, Kyoung Mu Lee ·

Recent research on super-resolution has progressed with the development of deep convolutional neural networks (DCNN). In particular, residual learning techniques exhibit improved performance... In this paper, we develop an enhanced deep super-resolution network (EDSR) with performance exceeding those of current state-of-the-art SR methods. The significant performance improvement of our model is due to optimization by removing unnecessary modules in conventional residual networks. The performance is further improved by expanding the model size while we stabilize the training procedure. We also propose a new multi-scale deep super-resolution system (MDSR) and training method, which can reconstruct high-resolution images of different upscaling factors in a single model. The proposed methods show superior performance over the state-of-the-art methods on benchmark datasets and prove its excellence by winning the NTIRE2017 Super-Resolution Challenge. read more

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Super-Resolution BSD100 - 4x upscaling EDSR PSNR 27.71 # 16
SSIM 0.7420 # 17
Image Super-Resolution FFHQ 1024 x 1024 - 4x upscaling EDSR FID 15.54 # 3
MS-SSIM 0.933 # 5
PSNR 28.34 # 4
SSIM 0.827 # 4
Image Super-Resolution FFHQ 256 x 256 - 4x upscaling EDSR FID 129.14 # 4
MS-SSIM 0.901 # 4
PSNR 22.47 # 6
SSIM 0.706 # 4
Image Super-Resolution FFHQ 512 x 512 - 4x upscaling EDSR PSNR 30.188 # 2
SSIM 0.824 # 3
MS-SSIM 0.961 # 2
LLE 2.003 # 1
FED 0.0843 # 2
FID 20.605 # 7
LPIPS 0.2475 # 7
NIQE 13.636 # 6
Image Super-Resolution Manga109 - 4x upscaling EDSR PSNR 31.02 # 14
SSIM 0.9148 # 14
Image Super-Resolution Set14 - 4x upscaling EDSR PSNR 28.80 # 17
SSIM 0.7876 # 18
Image Super-Resolution Set5 - 4x upscaling EDSR PSNR 32.46 # 17
SSIM 0.8968 # 21
Image Super-Resolution Urban100 - 4x upscaling EDSR PSNR 26.64 # 14
SSIM 0.8033 # 14

Methods


No methods listed for this paper. Add relevant methods here