GTA-Human is built by coordinating a group of computational workers that simultaneously play the popular video game Grand Theft Auto V (GTA-V). 1.4 million SMPL parametric labels automatically annotated in 20 thousand video sequences were put together in this dataset. Besides the scale, GTA-Human contains more than 600 subjects, 20,000 action clips, six major categories of locations, realistically placed cameras, subject-environment interactions, lighting conditions, and varied types of weather.
Large-scale: 1.4 million SMPL annotations
Diverse features: characters, actions, locations, camera places, interactions, lighting and weather.