小目标检测算法推荐

冷不防 2023-01-23 08:55 12阅读 0赞

1.YOLTv4构建于YOLT和SIMRDWN之上，并更新这些框架以使用YOLO的最高效版本YOLOv4。YOLTv4被设计用于在任意大的图像中检测航空或卫星图像中的目标，这些图像远远超过深度学习目标检测框架通常接收到的~600×600像素大小。

这个存储库是建立在AlexeyAB的YOLOv4实现令人印象深刻的工作之上的，与YOLOv3（在SIMRDWN中实现）相比，YOLOv4提高了速度和检测性能。我们使用YOLOv4

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2phY2tlMTIx_size_16_color_FFFFFF_t_70][]

YOLTv4 builds upon [YOLT][] and [SIMRDWN][], and updates these frameworks to use the most performant version of [YOLO][], [YOLOv4][]. YOLTv4 is designed to detect objects in aerial or satellite imagery in arbitrarily large images that far exceed the ~600×600 pixel size typically ingested by deep learning object detection frameworks.

This repository is built upon the impressive work of AlexeyAB's [YOLOv4][] implementation, which improves both speed and detection performance compared to YOLOv3 (which is implemented in SIMRDWN). We use YOLOv4 insead of "[YOLOv5][]", since YOLOv4 is endorsed by the original creators of YOLO, whereas "YOLOv5" is not; furthermore YOLOv4 appears to have superior performance.

Below, we provide examples of how to use this repository with the open-source [Rareplanes][] dataset.

[https://github.com/avanetten/yoltv4][https_github.com_avanetten_yoltv4]

2018年YOLT算法You only look twice，改变一下思维，对大分辨率图片先进行分割，变成一张张小图，再进行检测。

为了避免两张小图之间，一些目标正好被分割截断，所以两个小图之间设置overlap重叠区域，比如分割的小图是960\*960像素大小，则overlap可以设置为960\*20%=192像素。

每个小图检测完成后，再将所有的框放到大图上，对大图整体做一次nms操作，将重叠区域的很多重复框去除。(按照在大图上裁剪的位置，直接回归到大图即可。)  
坏处就是增加了计算量，我们选用更轻的网络来balance一下吧~

[https://blog.csdn.net/moxibingdao/article/details/105852664][https_blog.csdn.net_moxibingdao_article_details_105852664]

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2phY2tlMTIx_size_16_color_FFFFFF_t_70]: /images/20221004/f1c4f9537a39407397cbff09ae6db575.png
[YOLT]: https://github.com/avanetten/yolt
[SIMRDWN]: https://github.com/avanetten/simrdwn
[YOLO]: https://pjreddie.com/darknet/yolo/
[YOLOv4]: https://github.com/AlexeyAB/darknet
[YOLOv5]: https://github.com/ultralytics/yolov5
[Rareplanes]: https://registry.opendata.aws/rareplanes/
[https_github.com_avanetten_yoltv4]: https://github.com/avanetten/yoltv4
[https_blog.csdn.net_moxibingdao_article_details_105852664]: https://blog.csdn.net/moxibingdao/article/details/105852664