move-iit-hpe

Our new dataset, move-iit-hpe is a Human Pose Estimation (HPE) dataset and competition captured with event-cameras [ ]. The dataset tests, in particular, the ability to deal with moving event-camera with cluttered scene and large variations in dynamics.

To subscibe for the latest updates about the project, subscribe with this form: https://forms.iit.it/view.php?id=662625

The dataset consists of 11 subjects performing 17 actions. The subject is captured with 3 cameras: a stationary event-camera, a moving event-camera, and a moving rgb camera. Each action is repeated twice, with the moving camera moving “slow” and “fast”. A VICON motion capture system is used for joint ground-truth, using the standard 13 joint model [ ].

Camera	Version	Resolution
static event-camera	Prophesee ATIS EVK4	1280x720
dynamic event-camera	Prophesee ATIS EVK4	1280x720
rgb camera	real-sense V?	640x480

Why event-cameras?

explain what and why event-cameras. especially for HPE.

TO-DO

• logo/simple image for dataset
• diagram of layout of cameras and subject
• collect and media from stefano, yvonne, luna, gaurvi etc.
• example videos from each of the sensors
• upload sample dataset - hosting?
• instructions and script for loading data
• dockerfile and instructions for testing
• autoplay video and move to centre