Visual relationship detection based on bidirectional recurrent neural network

Visual relationship detection is a task aiming at mining the information of interactions between the paired objects in the image, describing the image in the form of (subject − predicate − object). Most of the previous works regard it as a pure classification problem by taking the integrated triplets as the label of the image; however, the numerous combinations of objects and the diversity of predicates are the tough challenges for these studies. Hence, we propose a deep model based on a modified bidirectional recurrent neural network (BRNN) to classify object and predict predicate simultaneously. By using the BRNN, the hidden information of the relationship in the image is extracted and a feature-infusion method is proposed. Additionally, we improve the existing works by introducing a paired non-maximum suppression method. The experiments show that our approach is competitive with the state-of-the-art works.