JARVIS (Joint Acquisition, Recording and Voxel based Inference System) aims to make neural network based 3D markerless pose estimation easy. From calibrating your camera system to final 3D predictions, we to provide easy to use software for every step along the way.

• Our AcquisitionTool allows you to record synchronised videos from multiple views at high FPS thanks to GPU accelerated online JPEG compression.
• The AnnotationTool is a convenient way to extract and annotate training sets for our state of the art 3D pose estimation network.
• HybridNet is our hybrid 2D- and 3D-CNN based network architecture that enables highly precise markerless motion capture - even in scenarios with heavy occlusion and complex natural environments.

Check out https://jarvis-mocap.github.io/jarvis-docs/ for more information about designing a 3D motion capture setup and our Getting Started Guide.

The toolbox was developed at the Neurobiology Lab of the German Primate Center (DPZ). If you have any questions or other inquiries related to JARVIS please contact:

Timo Hüser - @hueser_timo - [email protected]