Scale-recurrent Network for Deep Image Deblurring

In single image deblurring, the "coarse-to-fine" scheme, i.e. gradually restoring the sharp image on different resolutions in a pyramid, is very successful in both traditional optimization-based methods and recent neural-network-based approaches. In this paper, we investigate this strategy and propose a Scale-recurrent Network (SRN-DeblurNet) for this deblurring task. Compared with the many recent learning-based approaches in [25], it has a simpler network structure, a smaller number of parameters and is easier to train. We evaluate our method on large-scale deblurring datasets with complex motion. Results show that our method can produce better quality results than state-of-the-arts, both quantitatively and qualitatively.

PDF Abstract CVPR 2018 PDF CVPR 2018 Abstract

Results from the Paper


Ranked #2 on Image Deblurring on GoPro (Params (M) metric, using extra training data)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Image Deblurring GoPro SRN SSIM 0.9342 # 40
Params (M) 8.06 # 2
Deblurring GoPro SRN SSIM 0.9342 # 43
Deblurring HIDE (trained on GOPRO) SRN PSNR (sRGB) 28.36 # 24
SSIM (sRGB) 0.915 # 22
Params (M) 8.06 # 3
Deblurring RealBlur-J SRN SSIM (sRGB) 0.909 # 16
PSNR (sRGB) 31.38 # 16
Params(M) 8.06 # 2
Deblurring RealBlur-J (trained on GoPro) SRN PSNR (sRGB) 28.56 # 12
Deblurring RealBlur-R SRN PSNR (sRGB) 38.65 # 15
SSIM (sRGB) 0.965 # 14
Params 8.06 # 1
Deblurring RealBlur-R (trained on GoPro) SRN SSIM (sRGB) 0.947 # 11
Deblurring RSBlur SRN-Deblur Average PSNR 32.53 # 12
Image Relighting VIDIT’20 validation set SRN PSNR 16.94 # 5
SSIM 0.5660 # 5
LPIPS 0.4319 # 4
MPS 0.5670 # 5
Runtime(s) 0.87 # 5

Methods


No methods listed for this paper. Add relevant methods here