yolov8加入DCN

yolov8加入DCN（可变形卷积网络）的确可以在目标检测中取得较好的改进效果。DCN被称为目标检测改进网络结构的中涨点神器，对小目标检测效果尤其好。作者在DCN的基础上进行了改进，增加了调制模块和使用多个调制后的DCN模块，形成了DCN-v2的升级版本。调制模块除了学习偏移参数之外，还通过调制学习一个变化幅度，通过这个幅度来合理控制新采样点的偏移范围。经过调制后的单个DCN记为mDCN。yolov8加入DCN-v2后，在多个数据集上进行验证，mAP（平均精确度均值）有明显的提升，特别是对于小目标和遮挡物的检测精度。<span class="em">1</span><span class="em">2</span><span class="em">3</span>
#### 引用[.reference_title]
- *1* [YOLOv8添加DCNv3可变形卷积](https://blog.csdn.net/weixin_70423469/article/details/131702564)[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^v93^chatsearchT3_2"}}] [.reference_item style="max-width: 50%"]
- *2* *3* [涨点技巧：卷积变体DCNV2引入Yolov8，助力小目标涨点](https://blog.csdn.net/m0_63774211/article/details/130215147)[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^v93^chatsearchT3_2"}}] [.reference_item style="max-width: 50%"]
[ .reference_list ]