Skip to content
Qunliang Xing edited this page Nov 5, 2021 · 10 revisions

Please refer to this repository for details of image quality assessment.

Environment

  • Ubuntu 20.04.
  • Tesla V100-SXM2-32GB @ 1 card.
  • Intel (R) Xeon (R) Platinum 8163 CPUs @ 2.50 GHz.

Data-set

DIV2K, 0801-0900.png

Method

  • Non-blind @ fidelity: AR-CNN, DCAD, DnCNN
  • Non-blind @ realness: ESRGAN, MW-GAN
  • Blind @ fidelity: CBDNet, RBQE

Efficacy

metric PSNR SSIM MS-SSIM LPIPS BRISQUE NIQE PIQE
raw N/A N/A N/A N/A 23.952 2.828 29.830
QF=10 27.095 0.781 0.914 0.323 52.925 5.696 63.103
QF=20 29.588 0.851 0.954 0.207 46.181 4.366 51.781
QF=30 30.911 0.880 0.967 0.154 41.249 3.740 46.170
QF=40 31.791 0.896 0.974 0.125 37.835 3.338 43.081
QF=50 32.489 0.909 0.978 0.103 34.752 3.021 40.401
QP=42 29.717 0.838 0.949 0.240 42.209 3.526 58.456
QP=37 32.056 0.890 0.968 0.165 38.542 3.156 51.635
QP=32 34.688 0.929 0.981 0.101 34.884 2.894 44.769
QP=27 37.547 0.957 0.990 0.048 30.745 2.787 38.442
QP=22 40.247 0.975 0.994 0.018 27.637 2.799 33.349
QF=10 @ AR-CNN 28.408 0.821 0.938 0.286 43.117 3.890 64.589
QF=20 @ AR-CNN 30.869 0.878 0.966 0.198 41.012 3.701 57.314
QF=30 @ AR-CNN 32.190 0.902 0.975 0.156 39.009 3.519 53.306
QF=40 @ AR-CNN 33.051 0.915 0.980 0.134 38.800 3.462 50.548
QF=50 @ AR-CNN 33.656 0.924 0.983 0.112 34.584 3.296 45.804
QP=42 @ AR-CNN 30.164 0.847 0.953 0.237 40.204 3.588 64.165
QP=37 @ AR-CNN 32.554 0.897 0.971 0.166 37.515 3.330 56.333
QP=32 @ AR-CNN 35.177 0.934 0.983 0.104 34.707 3.079 48.120
QP=27 @ AR-CNN 38.072 0.961 0.991 0.051 31.534 2.863 40.982
QP=22 @ AR-CNN 40.675 0.977 0.995 0.018 27.211 2.765 32.950
QF=10 @ CBDNet 29.254 0.842 0.951 0.253 39.437 3.827 66.771
QF=20 @ CBDNet 31.689 0.893 0.973 0.178 38.281 3.615 59.643
QF=30 @ CBDNet 32.990 0.914 0.980 0.143 37.544 3.503 55.259
QF=40 @ CBDNet 33.841 0.925 0.984 0.121 36.561 3.373 51.990
QF=50 @ CBDNet 34.515 0.934 0.986 0.105 35.760 3.282 50.051
QP=42 @ CBDNet 30.586 0.855 0.957 0.228 41.664 3.652 66.354
QP=37 @ CBDNet 33.071 0.905 0.974 0.157 39.379 3.411 59.805
QP=32 @ CBDNet 35.742 0.940 0.985 0.100 36.846 3.181 52.282
QP=27 @ CBDNet 38.573 0.964 0.992 0.052 33.275 2.980 44.240
QP=22 @ CBDNet 41.224 0.979 0.995 0.022 29.215 2.878 38.084
QF=10 @ DCAD 28.695 0.829 0.943 0.268 39.478 3.778 66.236
QF=20 @ DCAD 31.073 0.882 0.968 0.194 39.297 3.657 58.488
QF=30 @ DCAD 32.424 0.905 0.977 0.154 38.352 3.531 54.959
QF=40 @ DCAD 33.282 0.918 0.981 0.131 37.842 3.416 52.062
QF=50 @ DCAD 33.989 0.928 0.984 0.113 36.096 3.309 49.360
QP=42 @ DCAD 30.258 0.849 0.954 0.237 41.357 3.634 65.185
QP=37 @ DCAD 32.658 0.899 0.972 0.165 38.632 3.392 57.931
QP=32 @ DCAD 35.382 0.936 0.984 0.107 36.647 3.196 51.888
QP=27 @ DCAD 38.291 0.962 0.991 0.055 33.303 3.002 43.626
QP=22 @ DCAD 41.049 0.978 0.995 0.021 28.578 2.810 36.923
QF=10 @ DnCNN 28.823 0.831 0.945 0.267 39.713 3.796 66.803
QF=20 @ DnCNN 31.177 0.883 0.969 0.189 37.613 3.631 57.862
QF=30 @ DnCNN 32.501 0.906 0.977 0.152 37.535 3.516 53.777
QF=40 @ DnCNN 33.331 0.918 0.981 0.128 36.863 3.371 50.626
QF=50 @ DnCNN 34.032 0.928 0.984 0.111 35.770 3.285 47.536
QP=42 @ DnCNN 30.275 0.849 0.954 0.239 41.100 3.623 65.063
QP=37 @ DnCNN 32.718 0.900 0.972 0.164 38.569 3.388 58.410
QP=32 @ DnCNN 35.382 0.936 0.984 0.107 36.425 3.201 51.437
QP=27 @ DnCNN 38.243 0.962 0.991 0.054 32.892 2.961 43.162
QP=22 @ DnCNN 41.087 0.978 0.995 0.021 28.496 2.812 36.721
QF=10 @ ESRGAN 27.379 0.794 0.934 0.174 20.515 2.555 22.682
QF=20 @ ESRGAN 29.399 0.843 0.959 0.108 18.478 2.798 22.539
QF=30 @ ESRGAN 31.003 0.876 0.971 0.078 19.179 3.114 21.816
QF=40 @ ESRGAN 31.982 0.893 0.976 0.060 22.475 2.855 27.467
QF=50 @ ESRGAN 32.564 0.903 0.980 0.049 21.706 2.666 29.103
QP=42 @ ESRGAN 28.980 0.806 0.941 0.151 22.460 2.599 26.996
QP=37 @ ESRGAN 31.262 0.865 0.964 0.097 19.197 2.779 25.380
QP=32 @ ESRGAN 33.977 0.909 0.979 0.055 18.962 3.143 24.758
QP=27 @ ESRGAN 36.934 0.951 0.989 0.026 22.374 2.778 28.484
QP=22 @ ESRGAN 39.985 0.971 0.994 0.011 21.667 3.252 25.345
QF=10 @ MW-GAN 25.079 0.683 0.904 0.231 28.579 5.684 21.821
QF=20 @ MW-GAN 28.205 0.802 0.951 0.139 19.568 4.444 14.219
QF=30 @ MW-GAN 29.681 0.840 0.965 0.103 19.934 4.269 16.890
QF=40 @ MW-GAN 30.896 0.864 0.971 0.086 20.903 4.272 18.008
QF=50 @ MW-GAN 31.227 0.869 0.975 0.065 21.795 4.707 20.789
QP=42 @ MW-GAN 28.206 0.782 0.937 0.174 19.143 3.751 19.796
QP=37 @ MW-GAN 30.669 0.849 0.961 0.109 21.738 3.539 24.014
QP=32 @ MW-GAN 33.450 0.903 0.978 0.065 21.969 3.620 24.799
QP=27 @ MW-GAN 36.580 0.941 0.988 0.030 19.938 3.239 23.365
QP=22 @ MW-GAN 39.631 0.968 0.993 0.012 21.124 2.851 26.642
QF=10 @ RBQE 29.307 0.842 0.952 0.252 40.720 3.863 68.466
QF=20 @ RBQE 31.720 0.893 0.973 0.177 38.403 3.658 60.865
QF=30 @ RBQE 33.040 0.914 0.980 0.141 38.108 3.552 56.509
QF=40 @ RBQE 33.899 0.925 0.984 0.120 37.164 3.389 53.527
QF=50 @ RBQE 34.572 0.934 0.986 0.104 35.780 3.302 51.134
QP=42 @ RBQE 30.679 0.857 0.958 0.227 41.833 3.678 66.517
QP=37 @ RBQE 33.157 0.906 0.975 0.157 39.765 3.445 60.448
QP=32 @ RBQE 35.819 0.940 0.985 0.101 37.195 3.223 53.033
QP=27 @ RBQE 38.631 0.964 0.992 0.052 33.919 3.017 45.211
QP=22 @ RBQE 41.291 0.979 0.995 0.021 29.421 2.886 37.943

Using the default NIQE model in MATLAB.

metric PSNR SSIM MS-SSIM LPIPS BRISQUE NIQE PIQE
raw N/A N/A N/A N/A 23.952 2.130 29.830
QF=10 27.095 0.781 0.914 0.323 52.925 5.055 63.103
QF=20 29.588 0.851 0.954 0.207 46.181 3.910 51.781
QF=30 30.911 0.880 0.967 0.154 41.249 3.370 46.170
QF=40 31.791 0.896 0.974 0.125 37.835 3.028 43.081
QF=50 32.489 0.909 0.978 0.103 34.752 2.780 40.401
QP=42 29.717 0.838 0.949 0.240 42.209 3.196 58.456
QP=37 32.056 0.890 0.968 0.165 38.542 2.888 51.635
QP=32 34.688 0.929 0.981 0.101 34.884 2.631 44.769
QP=27 37.547 0.957 0.990 0.048 30.745 2.449 38.442
QP=22 40.247 0.975 0.994 0.018 27.637 2.330 33.349
QF=10 @ AR-CNN 28.408 0.821 0.938 0.286 43.117 3.595 64.589
QF=20 @ AR-CNN 30.869 0.878 0.966 0.198 41.012 3.368 57.314
QF=30 @ AR-CNN 32.190 0.902 0.975 0.156 39.009 3.160 53.306
QF=40 @ AR-CNN 33.051 0.915 0.980 0.134 38.800 3.110 50.548
QF=50 @ AR-CNN 33.656 0.924 0.983 0.112 34.584 2.903 45.804
QP=42 @ AR-CNN 30.164 0.847 0.953 0.237 40.204 3.052 64.165
QP=37 @ AR-CNN 32.554 0.897 0.971 0.166 37.515 2.891 56.333
QP=32 @ AR-CNN 35.177 0.934 0.983 0.104 34.707 2.706 48.120
QP=27 @ AR-CNN 38.072 0.961 0.991 0.051 31.534 2.532 40.982
QP=22 @ AR-CNN 40.675 0.977 0.995 0.018 27.211 2.309 32.950
QF=10 @ CBDNet 29.254 0.842 0.951 0.253 39.437 3.325 66.771
QF=20 @ CBDNet 31.689 0.893 0.973 0.178 38.281 3.174 59.643
QF=30 @ CBDNet 32.990 0.914 0.980 0.143 37.544 3.064 55.259
QF=40 @ CBDNet 33.841 0.925 0.984 0.121 36.561 2.970 51.990
QF=50 @ CBDNet 34.515 0.934 0.986 0.105 35.760 2.909 50.051
QP=42 @ CBDNet 30.586 0.855 0.957 0.228 41.664 3.187 66.354
QP=37 @ CBDNet 33.071 0.905 0.974 0.157 39.379 2.993 59.805
QP=32 @ CBDNet 35.742 0.940 0.985 0.100 36.846 2.813 52.282
QP=27 @ CBDNet 38.573 0.964 0.992 0.052 33.275 2.600 44.240
QP=22 @ CBDNet 41.224 0.979 0.995 0.022 29.215 2.428 38.084
QF=10 @ DCAD 28.695 0.829 0.943 0.268 39.478 3.338 66.236
QF=20 @ DCAD 31.073 0.882 0.968 0.194 39.297 3.269 58.488
QF=30 @ DCAD 32.424 0.905 0.977 0.154 38.352 3.136 54.959
QF=40 @ DCAD 33.282 0.918 0.981 0.131 37.842 3.039 52.062
QF=50 @ DCAD 33.989 0.928 0.984 0.113 36.096 2.963 49.360
QP=42 @ DCAD 30.258 0.849 0.954 0.237 41.357 3.164 65.185
QP=37 @ DCAD 32.658 0.899 0.972 0.165 38.632 2.957 57.931
QP=32 @ DCAD 35.382 0.936 0.984 0.107 36.647 2.818 51.888
QP=27 @ DCAD 38.291 0.962 0.991 0.055 33.303 2.623 43.626
QP=22 @ DCAD 41.049 0.978 0.995 0.021 28.578 2.398 36.923
QF=10 @ DnCNN 28.823 0.831 0.945 0.267 39.713 3.343 66.803
QF=20 @ DnCNN 31.177 0.883 0.969 0.189 37.613 3.213 57.862
QF=30 @ DnCNN 32.501 0.906 0.977 0.152 37.535 3.117 53.777
QF=40 @ DnCNN 33.331 0.918 0.981 0.128 36.863 2.982 50.626
QF=50 @ DnCNN 34.032 0.928 0.984 0.111 35.770 2.957 47.536
QP=42 @ DnCNN 30.275 0.849 0.954 0.239 41.100 3.139 65.063
QP=37 @ DnCNN 32.718 0.900 0.972 0.164 38.569 2.951 58.410
QP=32 @ DnCNN 35.382 0.936 0.984 0.107 36.425 2.803 51.437
QP=27 @ DnCNN 38.243 0.962 0.991 0.054 32.892 2.601 43.162
QP=22 @ DnCNN 41.087 0.978 0.995 0.021 28.496 2.401 36.721
QF=10 @ ESRGAN 27.379 0.794 0.934 0.174 20.515 1.837 22.682
QF=20 @ ESRGAN 29.399 0.843 0.959 0.108 18.478 1.885 22.539
QF=30 @ ESRGAN 31.003 0.876 0.971 0.078 19.179 2.062 21.816
QF=40 @ ESRGAN 31.982 0.893 0.976 0.060 22.475 2.157 27.467
QF=50 @ ESRGAN 32.564 0.903 0.980 0.049 21.706 2.038 29.103
QP=42 @ ESRGAN 28.980 0.806 0.941 0.151 22.460 1.984 26.996
QP=37 @ ESRGAN 31.262 0.865 0.964 0.097 19.197 1.906 25.380
QP=32 @ ESRGAN 33.977 0.909 0.979 0.055 18.962 2.052 24.758
QP=27 @ ESRGAN 36.934 0.951 0.989 0.026 22.374 2.014 28.484
QP=22 @ ESRGAN 39.985 0.971 0.994 0.011 21.667 2.399 25.345
QF=10 @ MW-GAN 25.079 0.683 0.904 0.231 28.579 4.406 21.821
QF=20 @ MW-GAN 28.205 0.802 0.951 0.139 19.568 3.390 14.219
QF=30 @ MW-GAN 29.681 0.840 0.965 0.103 19.934 3.227 16.890
QF=40 @ MW-GAN 30.896 0.864 0.971 0.086 20.903 3.425 18.008
QF=50 @ MW-GAN 31.227 0.869 0.975 0.065 21.795 3.757 20.789
QP=42 @ MW-GAN 28.206 0.782 0.937 0.174 19.143 2.664 19.796
QP=37 @ MW-GAN 30.669 0.849 0.961 0.109 21.738 2.566 24.014
QP=32 @ MW-GAN 33.450 0.903 0.978 0.065 21.969 2.734 24.799
QP=27 @ MW-GAN 36.580 0.941 0.988 0.030 19.938 2.504 23.365
QP=22 @ MW-GAN 39.631 0.968 0.993 0.012 21.124 2.061 26.642
QF=10 @ RBQE 29.307 0.842 0.952 0.252 40.720 3.381 68.466
QF=20 @ RBQE 31.720 0.893 0.973 0.177 38.403 3.201 60.865
QF=30 @ RBQE 33.040 0.914 0.980 0.141 38.108 3.098 56.509
QF=40 @ RBQE 33.899 0.925 0.984 0.120 37.164 2.981 53.527
QF=50 @ RBQE 34.572 0.934 0.986 0.104 35.780 2.919 51.134
QP=42 @ RBQE 30.679 0.857 0.958 0.227 41.833 3.207 66.517
QP=37 @ RBQE 33.157 0.906 0.975 0.157 39.765 3.024 60.448
QP=32 @ RBQE 35.819 0.940 0.985 0.101 37.195 2.843 53.033
QP=27 @ RBQE 38.631 0.964 0.992 0.052 33.919 2.633 45.211
QP=22 @ RBQE 41.291 0.979 0.995 0.021 29.421 2.438 37.943

Using a NIQE model trained over the DIV2K data-set.

Efficiency

method params FPS
AR-CNN 20,099 32.2
DCAD 40,451 6.2
DnCNN 40,451 3.4
ESRGAN 1,451,011 @ gen, 987,617 @ dis 0.7
MW-GAN 59,808,707 @ gen, 4,098,051 @ dis 1.5
CBDNet 1,860,806 1.0
RBQE @ QF=10 (Dynamic) 9.2
RBQE @ QF=20 (Dynamic) 9.3
RBQE @ QF=30 (Dynamic) 10.0
RBQE @ QF=40 (Dynamic) 8.8
RBQE @ QF=50 (Dynamic) 7.7
RBQE @ QP=42 (Dynamic) 9.0
RBQE @ QP=37 (Dynamic) 11.2
RBQE @ QP=32 (Dynamic) 9.8
RBQE @ QP=27 (Dynamic) 9.4
RBQE @ QP=22 (Dynamic) 10.3
RBQE exit 0 1 2 3 4
params 298,898 743,594 1,409,672 2,333,996 3,553,430