Context Attention Module for Human Hand Detection

Human hand detection is still a challenging problem in a real world setting because the hand appearance features are less and varied form. So it is very important to make good use of context and enhance hand feature through each other. In this paper, we propose Context Attention Feature Pyramid Networks CA-FPN for human hand detection, where Context Attention Module (CAM) is introduced as a new component of the pyramid architecture. The CAM aims to: (1) capture context information and reserve local edge feature, (2) figure out the relative context for hands. The proposed model achieves state-of-the-art performance on two challenging hand detection datasets, i.e. the Oxford hand dataset and the Vision for Intelligent and Applications (VIVA) Challenge dataset. What's more, processing time of CA-FPN is about 8.5FPS on one TITAN X GPU, which could be applied for real time application.