Key points
- Noitom Robotics released the HiPHI-Hand dataset on Hugging Face.
- The dataset contains 11 hours of egocentric stereo video with marker-based hand ground truth.
- NVIDIA used HiPHI-Hand as one source of training data for its EgoHand model.
Context
HiPHI-Hand adds annotated egocentric hand footage to the resources available for developing and evaluating hand-aware vision systems. Its marker-based ground truth distinguishes tracked hand positions and movements, while its stereo format provides two synchronized views of each recorded scene.
What this means for you
Directly applicable
You can use the released dataset to train, test, or evaluate egocentric hand-perception systems, especially when working with stereo video and marker-based annotations.
Who benefits: Developers building egocentric vision, hand-tracking, or human-interaction models can directly evaluate the dataset or use it for experimental training.
Share article: