Multi-Level Network for High-Speed Multi-Person Pose Estimation
This addresses a hard problem in pose estimation for computer vision applications, but it appears incremental as it builds on existing refinement approaches.
The paper tackles the problem of left/right joint type discrimination in multi-person pose estimation by proposing a Multi-level Network (MLN) that aggregates features from multiple levels, achieving comparable performance to conventional methods with a runtime speed of 42.2 FPS.
In multi-person pose estimation, the left/right joint type discrimination is always a hard problem because of the similar appearance. Traditionally, we solve this problem by stacking multiple refinement modules to increase network's receptive fields and capture more global context, which can also increase a great amount of computation. In this paper, we propose a Multi-level Network (MLN) that learns to aggregate features from lower-level (left/right information), upper-level (localization information), joint-limb level (complementary information) and global-level (context) information for discrimination of joint type. Through feature reuse and its intra-relation, MLN can attain comparable performance to other conventional methods while runtime speed retains at 42.2 FPS.