Read about this work on “Pose Estimation on NVIDIA Jetson platforms using OpenPifPaf“.
Visit the GitHub Repository of this work to find out more.
Check out our new edge-device-friendly Pose Estimator, TinyPose.

Multi-person pose estimation on mobile and edge has a wide range of applications, including occupancy analytics, sports video analytics, activity recognition, motion tracking, and augmented reality. Bottom-up pose estimation methods like OpenPifPaf can provide accurate pose estimation for such applications. OpenPifPaf is a well-known pose estimation model that was developed at EPFL university in Switzerland.

A year ago, when we started to work with OpenPifPaf, we did not get great results when we ran them on Nvidia Jetson devices. The issue is deploying complex deep learning models on such devices with limited memory is challenging. In order to fix the performance issue, we needed to use inference optimization tools, such as TensorRT, to be able to run deep learning models on edge devices. We have recently released our model and our work on Neuralet, and our article is featured in the NVIDIA Community blog and Social Media. 

Need help with your ML projects? Contact us at hello@galliot.us.