yolov8head

YOLOv8中的Head部分有一些变化。在YOLOv8中，Head部分采用了解耦头（Decoupled Head）和Anchor Free策略，与之前的版本YOLOv5不同。YOLOv5使用了耦合头（Coupled Head）和Anchor Based策略。在YOLOv8中，不再有之前的Objectness分支，只有解耦的分类和回归分支。此外，由于采用了DFL（Dynamic Feature Learning）的思想，回归头的通道数也发生了变化，变成了4 * reg_max的形式。<span class="em">1</span><span class="em">2</span><span class="em">3</span>
#### 引用[.reference_title]
- *1* [目标检测算法——YOLOV8——算法详解](https://blog.csdn.net/u012863603/article/details/128816715)[target="_blank" data-report-click={"spm":"1018.2226.3001.9630","extra":{"utm_source":"vip_chatgpt_common_search_pc_result","utm_medium":"distribute.pc_search_result.none-task-cask-2~all~insert_cask~default-1-null.142^v92^chatsearchT0_1"}}] [.reference_item style="max-width: 50%"]
- *2* *3* [YOLOv8 预览与使用指南](https://blog.csdn.net/JasperXzy/article/details/128892683)[target="_blank" data-report-click={"spm":"1018.2226.3001.9630","extra":{"utm_source":"vip_chatgpt_common_search_pc_result","utm_medium":"distribute.pc_search_result.none-task-cask-2~all~insert_cask~default-1-null.142^v92^chatsearchT0_1"}}] [.reference_item style="max-width: 50%"]
[ .reference_list ]