复现deep_sort_yolov4--main.py

客官°小女子只卖身不卖艺 2023-02-23 08:22 146阅读 0赞

参考代码：[https://github.com/yehengchen/Object-Detection-and-Tracking/tree/master/OneStage/yolo/deep\_sort\_yolov4][https_github.com_yehengchen_Object-Detection-and-Tracking_tree_master_OneStage_yolo_deep_sort_yolov4]

代码作者分享yolov3与v4在townCentre.avi测试demo：

[https://www.bilibili.com/video/BV1Ug4y1i71w?from=search&seid=2376043687501739362][https_www.bilibili.com_video_BV1Ug4y1i71w_from_search_seid_2376043687501739362]

## 0.环境 ##

opencv-python==4.1.0.25
    Pillow
    scikit-learn==0.19.2
    numpy==1.15.0
    keras==2.2.4
    tensorflow-gpu==1.12.0
    imutils
    matplotlib

## 1.修改 ##

### 1.1 下载 ###

（1）下载代码作者自己训练的yolov4模型：[ \[yolo4\_weight.h5\] - w17w][_yolo4_weight.h5_ - w17w]

![20200706142421877.png][]

（2）下载TownCentre.avi，并将其放在./test\_video/TownCentreXVID.avi。

### 1.2 修改main.py ###

（1）注释\# from tensorflow.compat.v1 import InteractiveSession

line 26-33
    # from tensorflow.compat.v1 import InteractiveSession
    # config = tf.ConfigProto()
    # config.gpu_options.allow_growth = True
    # session = InteractiveSession(config=config)
    
    config = tf.ConfigProto()
    config.gpu_options.allow_growth = True
    session = tf.Session(config=config)

（2）由于在服务器中去测试的，所以关于opencv显示部分的注释掉：

line 165-167
    # cv2.namedWindow("YOLO4_Deep_SORT", 0);
    # cv2.resizeWindow('YOLO4_Deep_SORT', 1024, 768);
    # cv2.imshow('YOLO4_Deep_SORT', frame)
    
    line 201
    # cv2.destroyAllWindows()

（3）由于测试过程不显示图像了，所以使用打印输出看测试信息：

line 168-170
    print("Total Pedestrian Counter: "+str(count))
    print("Current Pedestrian Counter: "+str(i))
    print("FPS: %f"%(fps))

## 2.运行main.py ##

CUDA_VISIBLE_DEVICES=0 python3 main.py -c person

## 3.总结 ##

我测了下我自己的视频数据：在titan xp上，时间性能上FPS：5.9897与yolov3 FPS: 4.57相比，有一些优化。

## 参考 ##

[1.deep\_sort\_yolov4(yehengchen)][https_github.com_yehengchen_Object-Detection-and-Tracking_tree_master_OneStage_yolo_deep_sort_yolov4]

2.YOLOv4: Optimal Speed and Accuracy of Object Detection

3.Simple Online and Realtime Tracking with a Deep Association Metric

4.Deep Cosine Metric Learning for Person Re-identification

[https_github.com_yehengchen_Object-Detection-and-Tracking_tree_master_OneStage_yolo_deep_sort_yolov4]: https://github.com/yehengchen/Object-Detection-and-Tracking/tree/master/OneStage/yolo/deep_sort_yolov4
[https_www.bilibili.com_video_BV1Ug4y1i71w_from_search_seid_2376043687501739362]: https://www.bilibili.com/video/BV1Ug4y1i71w?from=search&seid=2376043687501739362
[_yolo4_weight.h5_ - w17w]: https://pan.baidu.com/s/1JuT4KCUFaE2Gvme0_S37DQ
[20200706142421877.png]: /images/20230209/4bbb7d179fbe40d7952c1b0ae140bed7.png