We evaluate our methods on two sets of images. We hand-picked a set of showcase photos of the most interesting portraits of historical figures. To conduct a fair and comprehensive user study, we also construct a diverse test benchmark, Historical Wiki Face Dataset, auto-selected by crawling Wikipedia. Our benchmark covers diverse styles and ethnic groups of important historical people.
In this supplementary material, we present a full table of comparisons with other baselines over on both the showcase set and the Historical Wiki Face Dataset. We also compare the restored antique photos with ground truth modern color photos. Then the PDF supplementary material describes our effects of Color Transfer Loss and implementation details.
