eduzhai > Applied Sciences > Engineering >

EfficientHRNet Efficient Scaling for Lightweight High-Resolution Multi-Person Pose Estimation

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 13 pages

Abstract: There is an increasing demand for lightweight multi-person pose estimationfor many emerging smart IoT applications. However, the existing algorithms tendto have large model sizes and intense computational requirements, making themill-suited for real-time applications and deployment on resource-constrainedhardware. Lightweight and real-time approaches are exceedingly rare and come atthe cost of inferior accuracy. In this paper, we present EfficientHRNet, afamily of lightweight multi-person human pose estimators that are able toperform in real-time on resource-constrained devices. By unifying recentadvances in model scaling with high-resolution feature representations,EfficientHRNet creates highly accurate models while reducing computation enoughto achieve real-time performance. The largest model is able to come within 4.4 accuracy of the current state-of-the-art, while having 1 3 the model size and1 6 the computation, achieving 23 FPS on Nvidia Jetson Xavier. Compared to thetop real-time approach, EfficientHRNet increases accuracy by 22 whileachieving similar FPS with 1 3 the power. At every level, EfficientHRNet provesto be more computationally efficient than other bottom-up 2D human poseestimation approaches, while achieving highly competitive accuracy.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×