Yolov5 advanced III training environment

Before training your own samples , First, it is necessary to debug yolov5 The training environment , Before we applied pycharm It was debugged , Actually provided yolov5 It can complete various functions in the command console .
First step , Clean up all folders , Leave only …_env Environment folder . That is to say, it is installed pytorch Environment . Unzip in the environment parallel directory yolov5-master Compressed package , Why decompress parallel directories , Because the default dataset Is in yolov5 Under the parallel directory of .

Just have the above two folders

Note the environment and current directory in the command console .

python detect.py --source data/images/zidane.jpg --weights yolov5s.pt

Execute the above order distinguish Zidane pictures . This will automatically download yolov5s.pt, If you want to use 5m 5l Just change the command , Enter the palace through the camera , perform

python detect.py --source 0

But pay attention to , Now this camera can't be turned off , Need to be in detect.py Add instructions to , Refer to advanced level 1 .
When the download is complete , The result of identification is ex? See the following figure for the table of contents
Let's test the training environment , If installed according to the advanced one pytorch Version is the latest version 11, Unfortunately, this version is different from yolov5 6.0 Not very compatible , It is recommended to reduce to a lower version pytorch
Execute the following command

pip install --upgrade --force-reinstall torch==1.8.0 torchvision==0.9.0 torchaudio==0.8.0

After downgrade , perform

python train.py --img 640 --batch 16 --epochs 5 --data ./data/coco128.yaml --cfg ./models/yolov5s.yaml --weights yolov5s.pt

Will automatically download coco128, Path and yolomaster parallel , This process still uses yolov5s Weight model , New... Will be generated after training last and best pt Weight model .
During the training, you can see gpu usage , We haven't tested yet gpu edition , Use cpu when gpu_mem Show 0

After training , Will be in runs/ Create one in the directory train Save the weight file in the folder . You can use the new weight file for image detection

python detect.py --weights runs/train/exp/weights/best.pt --source data/images/zidane.jpg```