We present MocapNET, a real-time method that estimates the 3D human pose directly in the popular Bio Vision Hierarchy (BVH) format, given estimations of the 2D body joints originating from monocular color images. Our contributions include: (a) A novel and compact 2D pose NSRM representation. (b) A human body orientation classifier and an ensembl…

Using a single RGB frame for real time 3D hand pose estimation in the wild

A Dockerfile for our WACV18 paper: Using a single RGB frame for real time 3D hand pose estimation in the wild.

Python bindings for the Openpose library

A ROS node for the MocapNET 3D Pose Estimator

Reading group material and links

Accurate Hand Keypoint Localization on Mobile Devices

3D Hand Tracking using input from a depth sensor.