Abstract | ||
---|---|---|
In this work we adapt multi-person pose estimation architecture to use it on edge devices. We follow the bottom-up approach from OpenPose (Cao et al., 2017), the winner of COCO 2016 Keypoints Challenge, because of its decent quality and robustness to number of people inside the frame. With proposed network design and optimized post-processing code the full solution runs at 28 frames per second (fps) on Intel (R) NUC 6i7KYB mini PC and 26 fps on Core i7-6850K CPU. The network model has 4.1M parameters and 9 billions floating-point operations (GFLOPs) complexity, which is just similar to 15% of the baseline 2-stage OpenPose with almost the same quality. The code and model are available as a part of Intel (R) OpenVINO (TM) Toolkit. |
Year | DOI | Venue |
---|---|---|
2018 | 10.5220/0007555407440748 | ICPRAM: PROCEEDINGS OF THE 8TH INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION APPLICATIONS AND METHODS |
Keywords | Field | DocType |
Human Pose Estimation, Keypoints, Joints, Bottom-up, OpenPose, Real-time | Network planning and design,Pattern recognition,FLOPS,Computer science,Pose,Robustness (computer science),Edge device,Computational science,Frame rate,Artificial intelligence,Unicode,Network model | Journal |
Volume | Citations | PageRank |
abs/1811.12004 | 2 | 0.40 |
References | Authors | |
5 | 1 |
Name | Order | Citations | PageRank |
---|---|---|---|
Daniil Osokin | 1 | 2 | 0.40 |