Zhipeng Fan

Logo

3rd Year Ph.D. student at NYU ECE; Enthusiastic about 3D Computer Vision and video understanding

Projects

Posts

Resume

View My GitHub Profile

3D Pose estimation for lymphedema patient therapy

Project description: In this on-going project (as part of the Optimal Lymph Flow system), we are looking into the possibility to develop an monocular camera based human pose estiamtion system for automatic evaluation of patients’ after surgery therapy exercises, which was previously realized with a Mircrosoft Kinect.

Preliminary evaluation with the VideoPose3D models by FAIR

We evaluated the performance of the temporal convolution model introduced in 3D human pose estimation in video with temporal convolutions and semi-supervised training on our pre-recorded lymphflow exercises. Some visualizations based on the VideoPose3D model are as followed. We modify the visualization codes to show the estiamted pose from 3 different angles: