Grasp UNderstanding (GUN-71) dataset

12,000 first-person RGB-D images of object manipulation scenes annotated using a taxonomy of 71 fine-grained grasps. (Rogez, Supancic and Ramanan)