Deeply-Recursive Convolutional Network for Image Super-Resolution

CVPR 2016  ·  Jiwon Kim, Jung Kwon Lee, Kyoung Mu Lee ·

We propose an image super-resolution method (SR) using a deeply-recursive convolutional network (DRCN). Our network has a very deep recursive layer (up to 16 recursions)... Increasing recursion depth can improve performance without introducing new parameters for additional convolutions. Albeit advantages, learning a DRCN is very hard with a standard gradient descent method due to exploding/vanishing gradients. To ease the difficulty of training, we propose two extensions: recursive-supervision and skip-connection. Our method outperforms previous methods by a large margin. read more

PDF Abstract CVPR 2016 PDF CVPR 2016 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Super-Resolution BSD100 - 2x upscaling DRCN [[Kim et al.2016b]] PSNR 31.85 # 15
Image Super-Resolution BSD100 - 4x upscaling DRCN PSNR 27.21 # 38
SSIM 0.7493 # 6
MOS 2.12 # 3
Image Super-Resolution Set14 - 2x upscaling DRCN [[Kim et al.2016b]] PSNR 33.04 # 14
Image Super-Resolution Set14 - 4x upscaling DRCN PSNR 28.02 # 38
SSIM 0.8074 # 4
MOS 2.84 # 3
Image Super-Resolution Set5 - 2x upscaling DRCN [[Kim et al.2016b]] PSNR 37.63 # 16
Image Super-Resolution Set5 - 4x upscaling DRCN PSNR 31.52 # 35
SSIM 0.8938 # 25
MOS 3.26 # 3
Image Super-Resolution Urban100 - 2x upscaling DRCN [[Kim et al.2016b]] PSNR 30.75 # 14

Methods


No methods listed for this paper. Add relevant methods here