[Deep Learning] Yolov8 tracks from 0 to 1, if it is doing counting, a simple batch

---

foreword

I have used yolov5 for nearly 2 years. Before that, I mainly did target detection, and also did yolov5 segmentation tasks and model conversion. Now I have new tasks. Time to play around with the new model.
The best way is to look at the documentation: https://docs.ultralytics.com/

It can be seen that compared with the past, yolov has changed from a target monitoring model to a comprehensive magic board, which can be an excellent choice for tasks including object inspection, tracking, instance segmentation, image classification and pose estimation.
Its pre-trained model can be downloaded here: https://github.com/ultralytics/ultralytics/blob/main/README.zh-CN.md
For example, monitoring:

segmentation:

classification:

pose estimation:

1. Task tracking

git:https://github.com/mikel-brostrom/yolov8_tracking
Real-time multi-object, segmentation and pose tracking using Yolov8 with DeepOCSORT and LightMBN

1.1 Build environment

# yolov8 现在格高了,封装成库了
pip install ultralytics
pip install lap filterpy easydict
pip install gdown

The installation requirement.txt has not been completed here, and it has a torch environment, which does not start from 0.

Download here, re-identify the weight: https://kaiyangzhou.github.io/deep-person-reid/MODEL_ZOO

and put it in the weights folder.

Just run the command:

$ python track.py --yolo-model yolov8n.pt      # bboxes only
                                 yolov8n-seg.pt  # bboxes + segmentation masks
                                 yolov8n-pose.pt # bboxes + pose estimation

This is the easiest way, everything else is the default

$ python track.py --source 0 --yolo-model yolov8n.pt --img 640
                                          yolov8s.tflite
                                          yolov8m.pt
                                          yolov8l.onnx 
                                          yolov8x.pt --img 1280
                                          ...

source 0 is the default webcam, computer camera

With a camera, the default is to have a graphics card. A picture takes 10ms, and the graphics card stands at 30% 12G 3070. It feels okay.
Because it can be tracked, it only needs to be commercialized and draw a line to count. Leave a message if you need it , I will implement it.
You can also add a few more parameters: --show --save will keep the file and display the recognition status.

–classes 16 17 can filter categories

2. Cross-camera tracking (advanced)

https://blog.csdn.net/qq_42312574/article/details/128880805
Cross-camera has always been the direction I want to try, and finally found the keyword: Multi-Target Multi-Camera Tracking (MTMC Tracking)
https://zhuanlan. zhihu.com/p/35391826, Great God Luo, but the age is a bit old, I guess there is a new sota now.

https://github.com/JunweiLiang/Object_Detection_Tracking base Tensorflow is cross-camera tracking, there are renderings
https://github.com/Jason-cs18/Awesome-Multi-Camera-Network A lot of learning materials are listed, no code
https://github.com/SurajDonthi/Multi-Camera-Person-Re-Identification/tree/master This is based on torch in 2021
https://github.com/cw1204772/AIC2018_iamai This is from 2018, and it also has code , seems to be tracking vehicles

Summarize

The author of yolov8 analyzes the content of yolov8
https://www.bilibili.com/video/BV17D4y1N7Zz/?spm_id_from=333.337.search-card.all.click&vd_source=3f7ae4b9d3a2d84bf24ff25f3294d107