Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

CVPR 2017 Christian LedigLucas TheisFerenc HuszarJose CaballeroAndrew CunninghamAlejandro AcostaAndrew AitkenAlykhan TejaniJohannes TotzZehan WangWenzhe Shi

Despite the breakthroughs in accuracy and speed of single image super-resolution using faster and deeper convolutional neural networks, one central problem remains largely unsolved: how do we recover the finer texture details when we super-resolve at large upscaling factors? The behavior of optimization-based super-resolution methods is principally driven by the choice of the objective function... (read more)

PDF Abstract

Code


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT LEADERBOARD
Image Super-Resolution BSD100 - 4x upscaling SRGAN PSNR 25.16 # 37
SSIM 0.6688 # 36
MOS 3.56 # 1
Image Super-Resolution BSD100 - 4x upscaling nearest neighbors PSNR 25.02 # 38
SSIM 0.6606 # 37
MOS 1.11 # 6
Image Super-Resolution BSD100 - 4x upscaling SRResNet PSNR 27.58 # 14
SSIM 0.7620 # 3
MOS 2.29 # 2
Image Super-Resolution BSD100 - 4x upscaling bicubic PSNR 25.94 # 35
SSIM 0.6935 # 34
MOS 1.47 # 5
Image Super-Resolution PIRM-test SRGAN NIQE 2.71 # 3
Image Super-Resolution Set14 - 4x upscaling bicubic PSNR 25.99 # 48
SSIM 0.7486 # 39
MOS 1.80 # 5
Image Super-Resolution Set14 - 4x upscaling SRResNet PSNR 28.49 # 19
SSIM 0.8184 # 2
MOS 2.98 # 2
Image Super-Resolution Set14 - 4x upscaling SRGAN PSNR 26.02 # 47
SSIM 0.7397 # 41
MOS 3.72 # 1
Image Super-Resolution Set14 - 4x upscaling nearest neighbors PSNR 24.64 # 50
SSIM 0.7100 # 42
MOS 1.20 # 6
Image Super-Resolution Set5 - 4x upscaling SRGAN PSNR 29.40 # 41
SSIM 0.8472 # 36
MOS 3.58 # 1
Image Super-Resolution Set5 - 4x upscaling nearest neighbors PSNR 26.26 # 46
SSIM 0.7552 # 40
MOS 1.28 # 6
Image Super-Resolution Set5 - 4x upscaling SRResNet PSNR 32.05 # 17
SSIM 0.9019 # 5
MOS 3.37 # 2
Image Super-Resolution Set5 - 4x upscaling bicubic PSNR 28.43 # 44
SSIM 0.8211 # 38
MOS 1.97 # 5

Results from Other Papers


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK SOURCE PAPER COMPARE
Image Super-Resolution FFHQ 1024 x 1024 - 4x upscaling SRGAN FID 60.67 # 7
MS-SSIM 0.807 # 7
PSNR 21.49 # 7
SSIM 0.515 # 7
Image Super-Resolution FFHQ 256 x 256 - 4x upscaling SRGAN FID 156.07 # 7
MS-SSIM 0.757 # 7
PSNR 17.57 # 7
SSIM 0.415 # 7
Image Super-Resolution VggFace2 - 8x upscaling SRGAN PSNR 23.01 # 3
Image Super-Resolution WebFace - 8x upscaling SRGAN PSNR 24.49 # 3