Realtime Multi-Person 2D Pose Estimation using ShuffleNet
Chen-zhi Guan · 2019
This paper analyzes the principles of the realtime multi-person 2D pose estimation, and then describes the design of the new method called Pose-ShuffleNet. The Pose-ShuffleNet is based on the building blocks of ShuffleNet unit and has the architecture of multi-stages and two branches jointly learning parts detection and parts association. The paper presents the comparison results among the method and other state-of-the-art methods, showing Pose-ShuffleNet more suitable for resource-constrained scenarios.