Object detection yolov5 open source project debugging

One 、 Open source code download

Project address ：https://github.com/ultralytics/yolov5

Two 、 Environment configuration

1. Load project

Decompress the downloaded file , open PyCharm, adopt File-Open, Select the project path to load the project .

2. Creating a virtual environment

1. Installation is required first Anaconda, And complete the configuration of some required environment variables . Then open the Anocanda Prompt, adopt conda create -n “name” python=“version” Creating a virtual environment （ The virtual environment is installed to by default C Plate , To modify the default path, refer to be based on Anaconda Module installation and precautions ）.
2. Install the corresponding module in the current virtual environment , The module information is in requirement.txt In file . stay Pycharm Open in Terminal, adopt pip intall -r requirement.txt install .
   
3. Get into Pycharm, adopt File-Settings-Project:yolov5-5.0-Python Interpreter Get into Python Interpreter configuration （ What I choose here is yolov5 Of 5.0 edition , You can download different versions as needed , So far it has arrived 6.1 edition ）, Add virtual environment of Python Interpreter path , Now the environment configuration is complete .
   

3、 ... and 、 be based on yolov5 The forecast

1. Prediction based on terminal command line

1. from https://github.com/ultralytics/yolov5/releases/tag/v6.1 Download a network weight file , And put it in the project directory , This project downloads yolo5x.pt.
2. stay Terminal Type in python detect.py --weights yolo5x.pt An error will be reported , The error information is as follows .
   Specific solutions can be referred to yolov5–Can’t get attribute ‘SPPF’ on
3. But after solving the above problems , An error message will appear during operation , The error information is as follows .
   
   Specific solutions can be referred to YOLOV5 There is a problem when training the model test ：AttributeError: ‘Upsample‘ object has no attribute ‘recompute_scale_factor‘ Solutions for
4. After solving the above problems, the code can run normally , The terminal displays as follows .
   

2. be based on Pycharm Prediction of integrated development environment

step 1、2、3 Are the same as those predicted based on the terminal command line . The different operations are as follows .

1. Click on the detect Icon , And select Edit Configurations.
   
2. stay Parameters Options --weights yolov5x.pt And Application .
   
3. function detect.py Results are displayed after .
   

3. Predicted results

Four 、 Code related explanations

'--weights'： Used to specify the weight parameters of the network , The default is yolov5s.pt（ The above project adopts yolov5x.pt, So it is necessary to specify again ）.
'--source'： Path of picture or video , That is, the path of network input .
'--img-size'： Specify the image size of the input network .
'--conf-thres'： The confidence threshold of the object , The target box below this threshold is not displayed .
'--iou-thres'：IOU threshold , For non maximum suppression . If the two check boxes IOU Greater than threshold , Then the detection box with higher reliability will be preserved . On the contrary, both detection boxes are reserved .
'--device'： use CPU still GPU operation .
'--view-img'： Display the results when running the code , among action Parameter set to store_true Indicates that if this parameter is specified, the result will be displayed （ There are also action Parameters , I won't go into that ）. For example, input at the terminal python detect.py --weights yolov5x.pt --view-img The running results will be displayed in real time during the running process .
'--save-txt'： Write the running results to txt In file .
'--save-conf': Save confidence .
'--nosave'： Whether to save the running results .
'--classes'： It can be used for category filtering , Show only targets in a specific category .
'--agnostic-nms'： Enhanced non maximum suppression .
'--augment'： An enhanced inference of the result .
'--project'： Save path of network input results .
'--name'： The name of the saved result .
'--exist-ok'： Whether the next run result overwrites the previous one .

notes ： In this paper, the reference object detection YOLOv5 Debugging and explanation of open source code project