Refinement Module based on Parse Graph of Feature Map for Human Pose Estimation

Abstract

The parse graph play a crucial role in enhancing the performance of human pose estimation (HPE). Its key advantage lies in its hierarchical structure, like a tree structure, and context relations among nodes, which enable more accurate for inference. To equip models with the advantage of parse graphs, many researchers predefine the parse graph of body structure for HPE. However, these frameworks struggle to adapt to instances that deviate from the predefined parse graph and they are often parameter-heavy. Unlike them, we view the feature map holistically, much like the human body. It can be optimized using parse graphs, where nodes' implicit feature representation boosts adaptability, avoiding rigid structural limitations. In this paper, we design the Refinement Module based on the Parse Graph of feature map (RMPG), which includes two stages: top-down decomposition and bottom-up combination. In the first stage, the feature map is constructed into a tree structure through recursive decomposition, with each node representing a sub-feature map, thereby achieving hierarchical modeling of features. In the second stage, context information is calculated and sub-feature maps with context are recursively connected to gradually build a refined feature map. Additionally, we design a hierarchical network with fewer parameters using multiple RMPG modules to model the context relations and hierarchies in the parse graph of body structure for HPE, some of which are supervised to obtain context relations among body parts. Our network achieves excellent results on multiple mainstream human pose datasets and the effectiveness of RMPG is proven on different methods. The code of RMPG will be open.

Tasks

Reproductions