Mo2Cap2

The Mo2Cap2 dataset is for training and evaluation of the egocentric 3D human body pose estimation method. The associated capture hardware is based on a novel lightweight setup that converts a standard baseball cap to a device for high-quality pose estimation based on a single cap-mounted fisheye camera. The training set contains 530,000 rendered images of human body with ground truth 2D and 3D annotation, which encompass around 3000 different actions and more than 700 different body textures. The test data contains more than 5000 real images captured with our cap-mounted hardware.