Light3DPose: Real-time Multi-Person 3D Pose Estimation from Multiple Views
CrossMind.ai logo
Details
Demo video of the method described in: Light3DPose: Real-time Multi-Person 3D Pose Estimation from Multiple Views Our approach can perform direct 3D pose estimation of multiple people from 1 up to 10 camera views at 6 FPS, processing on a single 1080Ti GPU. It is computationally independent of the number of people in the scene and exploits a very lightweight 2D backbone resulting in fast inference time. Arxiv: https://arxiv.org/pdf/2004.02688.pdf Authors: Alessio Elmi, Davide Mazzini and Pietro Tortella Checkout Technologies https://checkoutfree.it/

Comments
loading...