EgoHands is a dataset for recognizing complex hand gestures. The dataset contains 48 Google Glass videos containing complex first-person interactions between two people. The main goal of the dataset is to enable a better, data-driven understanding of hand gestures in first-person computer vision, which can semantically distinguish the observer's hands from other people's hands, as well as distinguish between left and right hands. The hand gestures of the actors are almost unconstrained as they freely engage in a range of interactions.

The dataset includes:

• High-quality, pixel-perfect hand images
• 15053 labeled hands data