Below we show results of all baseline models (before and after our inverse optimization scheme) discussed in our paper, over images from our WikiEarth benchmark. For each meta image, we sample random images (same across all methods). Each visualization shows the randomly-selected Internet image in the upper half and the rendered image from the reference model in the bottom half. We also show the ground-truth alignment (select GT below). Select the meta image and model that you would like to view.