AGORA stands for Avatars in Geography Optimized for Regression Analysis. It is a synthetic human dataset with high realism and very accurate ground truth. This dataset includes about 14k training images and 3k test images. It uses image-based lighting or rendered 3D environments so that each image includes 5-15 people. The AGORA dataset includes a total of 173k individual person crops, providing SPL/SMPL-X parameters and segmentation masks for each subject in the image.