Two-Stage Body-Part Attention Network for Detecting Human-Object Interactions

Human-Object Interaction (HOI) detection is a computer vision task that can help computers understand the scene of still images. While too many human-object pairs usually appear in images, which results in the complexity of the HOI detection network. In this paper, we used the two-stage framework in the HOI detection method to reduce complexity. Besides, based on the idea that the features of each body part have different degrees of importance for human-object interaction detection. We combined the body-part attention mechanism with the two-stage network to use relative information of human parts and object. In the test, we provided extensive experiments to compare the proposed network with the existing human-object interaction detection networks. The experiment on the V-COCO dataset shows that our approach compares favorably with the existing HOI detection methods.