Recently, Convolutional Neural Network (CNN) based models have achieved great success in Single Image SuperResolution (SISR). Owing to the strength of deep networks, these CNN models learn an effective nonlinear mapping from the low-resolution input image to the high-resolution target image, at the cost of requiring enormous parameters. This paper proposes a very deep CNN model (up to 52 convolutional layers) named Deep Recursive Residual Network (DRRN) that strives for deep yet concise networks. Specifically, residual learning is adopted, both in global and local manners, to mitigate the difficulty of training very deep networks; recursive learning is used to control the model parameters while increasing the depth. Extensive benchmark evaluation shows that DRRN significantly outperforms state of the art in SISR, while utilizing far fewer parameters.

Super Resolution

Results

Scale factor x2

x2_supp

Scale factor x3

x3_supp

Scale factor x4

x4_supp

Deep Recursive Residual Network Code

DRRN implementation may be downloaded here.

Publications

  • Image Super-Resolution via Deep Recursive Residual Network
    Ying Tai, Jian Yang, Xiaoming Liu
    In Proceeding of IEEE Computer Vision and Pattern Recognition (CVPR 2017), Honolulu, HI, Jul. 2017
    Bibtex | PDF | Project Webpage
  • @inproceedings{ image-super-resolution-via-deep-recursive-residual-network,
      author = { Ying Tai and Jian Yang and Xiaoming Liu },
      title = { Image Super-Resolution via Deep Recursive Residual Network },
      booktitle = { In Proceeding of IEEE Computer Vision and Pattern Recognition },
      address = { Honolulu, HI },
      month = { July },
      year = { 2017 },
    }