Images from the MIT-Adobe 5K dataset

First column shows the input images.

Second column shows the retouched images by Expert D.

Third column shows the predicted results for HDRNet.

Fourth and fifth column shows the results for 3 dimensions using PCA and VAE respectively.

Finally we show the results using our method in the last column.

You can toggle between the predicted and the target image by clicking on the images. Moreover, you can rearrange the images in increasing or decreasing order of the metrics.

ID Input Expert D HDRNet PSNR 20.429 FLIP 0.379 Delta E 2000 10.726 PCA PSNR 18.366 FLIP 0.404 Delta E 2000 11.414 VAE PSNR 28.525 FLIP 0.174 Delta E 2000 3.961 Ours PSNR 38.243 FLIP 0.099 Delta E 2000 1.815
000001 22.882 0.353 8.281 19.609 0.363 9.065 31.320 0.140 2.719 37.483 0.094 1.386
000002 21.326 0.343 8.799 21.174 0.330 7.282 29.220 0.179 3.576 37.403 0.088 1.614
000003 23.399 0.324 7.870 13.497 0.633 20.373 27.672 0.148 3.723 31.349 0.153 2.493
000004 18.433 0.355 10.152 20.194 0.332 8.072 26.827 0.188 4.525 37.123 0.074 1.445
000005 19.680 0.363 8.174 22.949 0.283 6.110 22.876 0.260 5.039 48.070 0.032 0.357
000006 26.901 0.239 4.367 17.791 0.566 12.005 28.018 0.151 3.502 34.549 0.143 2.195
000007 19.974 0.356 9.582 16.759 0.429 11.732 22.349 0.279 6.192 38.321 0.087 1.668
000008 23.674 0.276 5.909 20.095 0.430 10.574 30.054 0.144 2.980 36.865 0.110 2.532
000009 20.494 0.386 7.995 21.549 0.302 6.075 30.241 0.155 2.890 38.695 0.091 1.488
000010 13.703 0.616 22.450 17.303 0.456 12.762 23.541 0.278 6.371 38.241 0.114 2.230