图像 super-resolution restruction 的各种主流实现方式

文章源地址：https://github.com/huangzehao/Super-Resolution.Benckmark

但是从这里最后的​Quantitative comparisons可以看出，如果单从传统学习的角度，对于PSNR的提升，可能已经很难超越DL。当然有一些基于重建的方法可能会接近，但是时间消耗太大，只能学术玩玩。

A curated list of super-resolution resources and a benchmark for single image super-resolution algorithms.

See my implementated super-resolution algorithms:

• SRGAN
• VDSR
• CSCN

TODO

Build a benckmark like SelfExSR_Code

State-of-the-art algorithms

Classical Sparse Coding Method

• ScSR [Web]
• Image super-resolution as sparse representation of raw image patches (CVPR2008), Jianchao Yang et al.
• Image super-resolution via sparse representation (TIP2010), Jianchao Yang et al.
• Coupled dictionary training for image super-resolution (TIP2011), Jianchao Yang et al.

Anchored Neighborhood Regression Method

• ANR [Web]
• Anchored Neighborhood Regression for Fast Example-Based Super-Resolution (ICCV2013), Radu Timofte et al.
• A+ [Web]
• A+: Adjusted Anchored Neighborhood Regression for Fast Super-Resolution (ACCV2014), Radu Timofte et al.
• IA [Web]
• Seven ways to improve example-based single image super resolution (CVPR2016), Radu Timofte et al.

Self-Exemplars

• SelfExSR [Web]
• Single Image Super-Resolution from Transformed Self-Exemplars (CVPR2015), Jia-Bin Huang et al.

Bayes

• NBSRF [Web]
• Naive Bayes Super-Resolution Forest (ICCV2015), Jordi Salvador et al.

Deep Learning Method

• SRCNN [Web] [waifu2x by nagadomi]
• Image Super-Resolution Using Deep Convolutional Networks (ECCV2014), Chao Dong et al.
• Image Super-Resolution Using Deep Convolutional Networks (TPAMI2015), Chao Dong et al.
• CSCN [Web]
• Deep Networks for Image Super-Resolution with Sparse Prior (ICCV2015), Zhaowen Wang et al.
• Robust Single Image Super-Resolution via Deep Networks with Sparse Prior (TIP2016), Ding Liu et al.
• VDSR [Web] [Unofficial Implementation in Caffe]
• Accurate Image Super-Resolution Using Very Deep Convolutional Networks (CVPR2016), Jiwon Kim et al.
• DRCN [Web]
• Deeply-Recursive Convolutional Network for Image Super-Resolution (CVPR2016), Jiwon Kim et al.
• ESPCN [PDF]
• Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network (CVPR2016), Wenzhe Shi et al.
• Is the deconvolution layer the same as a convolutional layer? [PDF]
• Checkerboard artifact free sub-pixel convolution [PDF]
• FSRCNN [Web]
• Acclerating the Super-Resolution Convolutional Neural Network (ECCV2016), Dong Chao et al.
• LapSRN [Web]
• Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution (CVPR 2017), Wei-Sheng Lai et al.
• EDSR [PDF]
• Enhanced Deep Residual Networks for Single Image Super-Resolution (Winner of NTIRE2017 Super-Resolution Challenge), Bee Lim et al.

Perceptual Loss and GAN

• Perceptual Loss [PDF]
• Perceptual Losses for Real-Time Style Transfer and Super-Resolution (ECCV2016), Justin Johnson et al.
• SRGAN [PDF]
• Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network (CVPR2017), Christian Ledig et al.
• AffGAN [PDF]
• AMORTISED MAP INFERENCE FOR IMAGE SUPER-RESOLUTION (ICLR2017), Casper Kaae Sønderby et al.
• EnhanceNet [PDF]
• EnhanceNet: Single Image Super-Resolution through Automated Texture Synthesis, Mehdi S. M. Sajjadi et al.
• neural-enchance [Github]

Video SR

• VESPCN [PDF]
• Real-Time Video Super-Resolution with Spatio-Temporal Networks and Motion Compensation (CVPR2017), Jose Caballero et al.

Dicussion

Deconvolution and Sub-Pixel Convolution

• Deconvolution and Checkerboard Artifacts
• SubPixel

Datasets

Test Dataset	Image source
Set 5	Bevilacqua et al. BMVC 2012
Set 14	Zeyde et al. LNCS 2010
BSD 100	Martin et al. ICCV 2001
Urban 100	Huang et al. CVPR 2015

Train Dataset	Image source
Yang 91	Yang et al. CVPR 2008
BSD 200	Martin et al. ICCV 2001
General 100	Dong et al. ECCV 2016
ImageNet	Olga Russakovsky et al. IJCV 2015
COCO	Tsung-Yi Lin et al. ECCV 2014

Quantitative comparisons

Results from papers of VDSR, DRCN, CSCN and IA.

Note: IA use enchanced prediction trick to improve result.

Results on Set 5

Scale	Bicubic	A+	SRCNN	SelfExSR	CSCN	VDSR	DRCN	IA
2x - PSNR/SSIM	33.66/0.9929	36.54/0.9544	36.66/0.9542	36.49/0.9537	36.93/0.9552	37.53/0.9587	37.63/0.9588	37.39/
3x - PSNR/SSIM	30.39/0.8682	32.59/0.9088	32.75/0.9090	32.58/0.9093	33.10/0.9144	33.66/0.9213	33.82/0.9226	33.46/
4x - PSNR/SSIM	28.42/0.8104	30.28/0.8603	30.48/0.8628	30.31/0.8619	30.86/0.8732	31.35/0.8838	31.53/0.8854	31.10/

Results on Set 14

Scale	Bicubic	A+	SRCNN	SelfExSR	CSCN	VDSR	DRCN	IA
2x - PSNR/SSIM	30.24/0.8688	32.28/0.9056	32.42/0.9063	32.22/0.9034	32.56/0.9074	33.03/0.9124	33.04/0.9118	32.87/
3x - PSNR/SSIM	27.55/0.7742	29.13/0.8188	29.28/0.8209	29.16/0.8196	29.41/0.8238	29.77/0.8314	29.76/0.8311	29.69/
4x - PSNR/SSIM	26.00/0.7027	27.32/0.7491	27.49/0.7503	27.40/0.7518	27.64/0.7587	28.01/0.7674	28.02/0.7670	27.88/

Results on BSD 100

Scale	Bicubic	A+	SRCNN	SelfExSR	CSCN	VDSR	DRCN	IA
2x - PSNR/SSIM	29.56/0.8431	31.21/0.8863	31.36/0.8879	31.18/0.8855	31.40/0.8884	31.90/0.8960	31.85/0.8942	31.79/
3x - PSNR/SSIM	27.21/0.7385	28.29/0.7835	28.41/0.7863	28.29/0.7840	28.50/0.7885	28.82/0.7976	28.80/0.7963	28.76/
4x - PSNR/SSIM	25.96/0.6675	26.82/0.7087	26.90/0.7101	26.84/0.7106	27.03/0.7161	27.29/0.7251	27.23/0.7233	27.25/