A Target Detection Method Combining Location Information and Knowledge Distillation

Target detection is one of the most important tasks in the field of computer vision to classify and locate targets in images or videos. This paper proposes a new algorithm for extracting feature knowledge. Designed to combine location information and knowledge distillation, extracting classification knowledge and localization knowledge separately on feature maps. Select the distillation area based on location information and decouple the feature map into the classification head and localization head, respectively. Perform different knowledge distillation and positional distillation on classification knowledge and localization knowledge, and design an overall distillation loss function. After experimental verification, while ensuring real-time performance, the average accuracy of this method on the COCO dataset has improved by 8.2% compared to the YOLOv5 algorithm.