EgoDexter Pose Estimation Dataset

The EgoDexter dataset provides 2D and 3D pose annotations for 3,190 frames of 4 test video sequences. These videos are shot from the self-perspective with a somatosensory camera and contain cluttered backgrounds, fast camera motions, and complex interactions with various objects. Among the 3,190 frames, 1,485 frames have fingertip positions manually annotated.