Name
Abstract | ||
|---|---|---|
Most object detection methods learn regional features in visual space while neglecting the relationship between targets, limiting their performance. We introduce the Graph Relation Decision Network (GRDN) to handle this problem, which includes the graph decision network, decision coefficient, and step-wise relation deduction module (SWRD). Among them, graph decision network contains edge decision network and node decision network. To obtain the implicit relationship between the labels in the dataset, we employ a data-driven technique. Through a graph decision network, the implicit relationship generates a dynamic graph, which can adaptively strengthen and enrich the relationship between the objects. We re-encode the dynamic graph. This process introduces decision coefficient, which can enhance semantic information. In the SWRD, we use semantic information to guide the basic visual features to prevent the visual features from being distracted. Experiments on the COCO dataset show that GRDN has improved performance in object detection and instance segmentation. |
Year | DOI | Venue |
|---|---|---|
2022 | 10.1109/ICME52920.2022.9859959 | 2022 IEEE International Conference on Multimedia and Expo (ICME) |
Keywords | DocType | ISSN |
Object Detection,Instance Segmentation,Relation Mining,Semantic Relation,Graph Neural Network | Conference | 1945-7871 |
ISBN | Citations | PageRank |
978-1-6654-8564-7 | 0 | 0.34 |
References | Authors | |
9 | 3 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Xiwei Yang | 1 | 0 | 0.34 |
| Xinfang Zhong | 2 | 0 | 0.34 |
| Zhixin Li | 3 | 12 | 19.62 |