YOLOv8 Detection Head Improvements for FPGA Deployments

Object detection is a core topic in the field of computer vision, which is widely used in industrial quality inspection, video surveillance, UAV target recognition and other scenes. These applications require extremely high speed of object detection. However, the existing advanced object detection networks often face computational bottlenecks. In order to solve this problem, this paper proposes a FPGA-based YOLOv8 target detection network data sharing detection head. By introducing a shared data module into the decoupling head, the design not only effectively reduces the amount of network computation, but also improves the application value of target detection technology, and opens up new possibilities for FPGA in high performance computing.