This paper introduces SURREAL (Synthetic hUmans foR REAL tasks), a large-scale dataset of synthetically generated but realistic images of people, rendered from 3D sequences of human motion capture data. The dataset includes over 6 million frames with ground truth pose, depth maps, and segmentation masks. The authors demonstrate that CNNs trained on this synthetic dataset can accurately estimate human depth and segment human parts in real RGB images. The paper also presents a detailed pipeline for generating synthetic data, including the use of a 3D human body model (SMPL), random sampling of body shapes, poses, textures, lighting, and backgrounds. The synthetic data is used to train CNNs for human body part segmentation and depth estimation, with experiments showing significant improvements over models trained only on real data. The results highlight the potential of using synthetic data for advancing person analysis tasks, particularly in terms of cost and scale.This paper introduces SURREAL (Synthetic hUmans foR REAL tasks), a large-scale dataset of synthetically generated but realistic images of people, rendered from 3D sequences of human motion capture data. The dataset includes over 6 million frames with ground truth pose, depth maps, and segmentation masks. The authors demonstrate that CNNs trained on this synthetic dataset can accurately estimate human depth and segment human parts in real RGB images. The paper also presents a detailed pipeline for generating synthetic data, including the use of a 3D human body model (SMPL), random sampling of body shapes, poses, textures, lighting, and backgrounds. The synthetic data is used to train CNNs for human body part segmentation and depth estimation, with experiments showing significant improvements over models trained only on real data. The results highlight the potential of using synthetic data for advancing person analysis tasks, particularly in terms of cost and scale.