Knowledge Graph-Guided Object Detection with Semantic Distance Network

This paper discusses the limitations of current object detection methods that only identify individual objects without considering their relationships with each other. To address this issue, the authors introduce a new approach called the Knowledge Graph-Guided Semantic Distance Network (KGSDN). This method incorporates a knowledge graph to provide semantic contextual cues, resulting in better object detection accuracy. The KGSDN framework combines the knowledge graph with the object detection network and uses an attention-based network to measure the semantic distance between objects. By updating the conditional object probability of each bounding box, the joint probability of all objects in an image is obtained. Experimental results indicate that this approach can enhance the performance of deep learning-based object detection methods.