https://github.com/quantumxiaol/yolov8-small-target-detection

基于yolov8实现小目标检测，在NWPU VHR-10和DOTA上测试
https://github.com/quantumxiaol/yolov8-small-target-detection

Last synced: 2 months ago
JSON representation

基于yolov8实现小目标检测，在NWPU VHR-10和DOTA上测试

• Host: GitHub
• URL: https://github.com/quantumxiaol/yolov8-small-target-detection
• Owner: quantumxiaol
• Created: 2024-05-22T11:34:52.000Z (9 months ago)
• Default Branch: main
• Last Pushed: 2024-05-22T12:35:45.000Z (9 months ago)
• Last Synced: 2024-05-22T12:46:52.416Z (9 months ago)
• Language: Python
• Size: 24.1 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-yolo-object-detection - quantumxiaol/yolov8-small-target-detection - small-target-detection?style=social"/> : 基于yolov8实现小目标检测，在NWPU VHR-10和DOTA上测试。 (Applications)
• awesome-yolo-object-detection - quantumxiaol/yolov8-small-target-detection - small-target-detection?style=social"/> : 基于yolov8实现小目标检测，在NWPU VHR-10和DOTA上测试。 (Applications)

README

        
# yolov8-small-target-detection

基于yolov8实现小目标检测，在NWPU VHR-10和DOTA上测试

使用Gradio-YOLOv8-Det进行可视化

yolov8  https://github.com/ultralytics/ultralytics

Gradio-YOLOv8-Det  https://gitee.com/CV_Lab/gradio-yolov8-det

可视化需要执行gradio_yolov8_det下的gradio_yolov8_det_v2.py。

根据要求修改/model_config/model_name_all.yaml以添加自己的模型权值

修改/cls_name/cls_name_zh.yaml以修改目标检测的标签值

文件结构

    |- yolov8 解压yolov8源代码

    

        |-datasets 储存数据集

    

               |--DOTAs

    

               |--NWPUVHR

    

        |-gradio_yolov8_det 结果可视化

    

        |-yolov8fornwpuvhr.pt  在NWPU VHR-10上训练的一个权值

    

        |-yolov8.yaml  配置网络，修改nc

    

        |-train.py  在NWPU VHR-10训练模型

        

        |-train_DOTAs.py  在DOTA训练模型

        

        |-pred.py  评估模型

    

        github.com/ultralytics/ultralytics

    

        |-docker

    

        |-docs

    

        |-docker

    

        |-examples

    

        |-ultralytics

可视化示意

![gradio_yolov8_det](https://github.com/quantumxiaol/yolov8-small-target-detection/blob/main/png/gradio_yolov8_det_examples.png "gradio_yolov8_det")

原图

![gradio_yolov8_det](https://github.com/quantumxiaol/yolov8-small-target-detection/blob/main/png/img_nwpu_test_raw.jpg "gradio_yolov8_det")

检测结果

![gradio_yolov8_det](https://github.com/quantumxiaol/yolov8-small-target-detection/blob/main/png/img_nwpu_test_det.jpg "gradio_yolov8_det")

在NWPU VHR-10数据集表现

![NWPU val_pred](https://github.com/quantumxiaol/yolov8-small-target-detection/blob/main/png/img_nwpu_val_pred.jpg "NWPU val_pred")

原标签

![NWPU val_labels](https://github.com/quantumxiaol/yolov8-small-target-detection/blob/main/png/img_nwpu_val_labels.jpg "NWPU val_labels")

在DOTA数据集表现

![DOTA val_pred](https://github.com/quantumxiaol/yolov8-small-target-detection/blob/main/png/img_dota_val_batch0_pred.jpg "DOTA val_pred")

原标签

![DOTA val_labels](https://github.com/quantumxiaol/yolov8-small-target-detection/blob/main/png/img_dota_val_batch0_labels.jpg "DOTA val_labels")

目前来看在DOTA上表现不好，这可能是由于在DOTA上进行标签转换时没有处理好旋转框，导致模型不能很好的学到较小的目标。

下面是修改了标签计算方法的训练结果。

![DOTA val_pred](https://github.com/quantumxiaol/yolov8-small-target-detection/blob/main/png/img_dota_val_batch1_pred_v2.jpg "DOTA val_pred")

原标签

![DOTA val_labels](https://github.com/quantumxiaol/yolov8-small-target-detection/blob/main/png/img_dota_val_batch1_labels_v2.jpg "DOTA val_labels")

Gradio-YOLOv8-Det的原作者  曾逸夫, (2024) Gradio YOLOv8 Det (Version 2.1.0).https://gitee.com/CV_Lab/gradio-yolov8-det.git.