Deep Back-Projection Networks for Single Image Super-resolution

4 Apr 2019  ·  Muhammad Haris, Greg Shakhnarovich, Norimichi Ukita ·

Previous feed-forward architectures of recently proposed deep super-resolution networks learn the features of low-resolution inputs and the non-linear mapping from those to a high-resolution output. However, this approach does not fully address the mutual dependencies of low- and high-resolution images. We propose Deep Back-Projection Networks (DBPN), the winner of two image super-resolution challenges (NTIRE2018 and PIRM2018), that exploit iterative up- and down-sampling layers. These layers are formed as a unit providing an error feedback mechanism for projection errors. We construct mutually-connected up- and down-sampling units each of which represents different types of low- and high-resolution components. We also show that extending this idea to demonstrate a new insight towards more efficient network design substantially, such as parameter sharing on the projection module and transition layer on projection step. The experimental results yield superior results and in particular establishing new state-of-the-art results across multiple data sets, especially for large scaling factors such as 8x.

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Super-Resolution BSDS100 - 2x upscaling DBPN-RES-MR64-3 PSNR 32.31 # 1
SSIM 0.901 # 1
Image Super-Resolution BSDS100 - 4x upscaling DBPN-RES-MR64-3 PSNR 27.82 # 1
SSIM 0.744 # 1
Image Super-Resolution BSDS100 - 8x upscaling DBPN-RES-MR64-3 PSNR 25.05 # 1
SSIM 0.607 # 1
Image Super-Resolution Manga109 - 2x upscaling DBPN-RES-MR64-3 PSNR 39.28 # 5
SSIM 0.977 # 6
Image Super-Resolution Manga109 - 4x upscaling DBPN-RES-MR64-3 PSNR 31.74 # 4
SSIM 0.921 # 5
Image Super-Resolution Manga109 - 8x upscaling DBPN-RES-MR64-3 PSNR 25.71 # 1
SSIM 0.813 # 1
Image Super-Resolution Set14 - 2x upscaling DBPN-RES-MR64-3 PSNR 34.09 # 6
SSIM 0.921 # 6
Image Super-Resolution Set14 - 4x upscaling DBPN-RES-MR64-3 PSNR 29.03 # 4
SSIM 0.791 # 11
Image Super-Resolution Set14 - 8x upscaling DBPN-RES-MR64-3 PSNR 25.41 # 1
SSIM 0.657 # 1
Image Super-Resolution Set5 - 2x upscaling DBPN-RES-MR64-3 PSNR 38.08 # 9
SSIM 0.96 # 7
Image Super-Resolution Set5 - 4x upscaling DBPN-RES-MR64-3 PSNR 32.65 # 11
SSIM 0.899 # 16
Image Super-Resolution Set5 - 8x upscaling DBPN-RES-MR64-3 PSNR 27.51 # 2
SSIM 0.793 # 2
Image Super-Resolution Urban100 - 2x upscaling DBPN-RES-MR64-3 PSNR 32.92 # 7
SSIM 0.935 # 6
Image Super-Resolution Urban100 - 4x upscaling DBPN-RES-MR64-3 PSNR 27.08 # 8
SSIM 0.814 # 10
Image Super-Resolution Urban100 - 8x upscaling DBPN-RES-MR64-3 PSNR 23.2 # 2
SSIM 0.652 # 2

Methods


No methods listed for this paper. Add relevant methods here