Yolo7 mask recognition practice

1. Make datasets

1.1 Install the dataset creation tool

Execute the following command to install the dataset creation tool labelimg

pip install labelimg

1.2 Run the dataset creation tool

Carry out orders labelimg, Run the dataset creation tool

labelimg The tool interface is as follows :

1.3 Make datasets

1. Click on Open Dir Button , Open the picture file to be made （ Try to place the picture in images Under the table of contents , It is convenient to split the data set later ）
   

2. Click the mode switch button on the left , Change the data pattern from PascalVOC Switch to YOLO Pattern
   

3. draw box box , And indicate the label
   Click on createBox Button , Draw a rectangular border , And set the label text （ No mask ）.
   

4. Click on Save Button , Save the sample label information （ Try to place the label file in labels Under the folder , Facilitate data segmentation later ）
   

5. Click on Next Image Button , Process the next image （ With mask ）
   
   After clicking on save Button , Save the sample label information （ Try to save the label information in labels Under the folder ）

6. Process the pictures one by one , You can complete the data set production

2. Using data sets

2.1 Download the mask data set

If you don't want to make data sets , You can download the mask data set ( Click to download , password ：sv74), And extract the data set to the root directory of the project （mask Catalog data sets ）.

2.2 Split the dataset

Download dataset segmentation code ( Click to download ), Put it on mask Under the table of contents , perform data_split.py, Divide the data set into training data set and verification data set . After split , The data set directory structure is as follows ：

You can adjust data_split.py Medium TRAIN_RATIO = 80, To indicate the proportion of training data set and validation data .

2.3 Dataset configuration

Copy data In the catalog coco.yaml, Save as mask.yaml, And modify the content . The revised content is as follows ：

train: ./mask/train.txt 
val: ./mask/val.txt 

nc: 2

names: [ 'no-mask', 'mask' ]

3. Training models

3.1 Download pre training model

Download pre training model ( Click to download ), Put it in the project root directory .

3.2 Modify the training code

A copy of train.py, And rename to train_mask.py.

open train_mask.py, Modify the following configurations ：

1. Pre training model name

parser.add_argument(‘–weights’, type=str, default=‘yolov7_training.pt’, help=‘initial weights path’)

2. Dataset path

parser.add_argument(‘–data’, type=str, default=‘data/mask.yaml’, help=‘data.yaml path’)

3. Number of training rounds ( Default 300, It took a long time , It may take hours or even days )

parser.add_argument(‘–epochs’, type=int, default=100)

4. Modify the model output path

parser.add_argument(‘–project’, default=‘runs/train/mask’, help=‘save to project/name’)

5. Adjust the number of pictures processed in each batch （ If the memory of the graphics card is small , It is recommended to change the value to 2 perhaps 4）

parser.add_argument(‘–batch-size’, type=int, default=2, help=‘total batch size for all GPUs’)

3.3 Training models

function train_mask.py, Model training . About 20 Round left and right , The accuracy of model recognition can reach 95% above .

runs\train\mask\exp11\weights\best.pt It is the best model parameter .

Can be best.pt Copy to the project root directory , And rename to yolov7_mask.pt, Convenient for later use .

4. Model test

4.1 Copy test code

Copy the test code file test.py, And rename to test_mask.py

4.2 Modify the test code

Modify the model parameter file

parser.add_argument(‘–weights’, nargs=‘+’, type=str, default=‘yolov7_mask.pt’, help=‘model.pt path(s)’)

Modify dataset path

parser.add_argument(‘–data’, type=str, default=‘data/mask.yaml’, help=‘*.data path’)

Modify the test item name

parser.add_argument(‘–project’, default=‘runs/test/mask’, help=‘save to project/name’)

4.3 Run test code

Run code test_mask.py, The running log is as follows ：

The operation results are as follows （ stay runs/test/mask/exp Under the table of contents ）：

5. problem

5.1 error ：FileNotFoundError: [Errno 2] No such file or directory

If pip install When installing the software , Tips FileNotFoundError: [Errno 2] No such file or directory error , You need to execute the following command to update pip And installation tools .

pip install --upgrade setuptools
python -m pip install --upgrade pip

5.2 error ：ModuleNotFoundError: No module named ‘tensorboard’

If something goes wrong ：ModuleNotFoundError: No module named ‘tensorboard’, execute pip install tensorboard Install this class library .

5.3 error ：If reserved memory is >> allocated memory

If you run out of memory , adjustment batch-size that will do , The adjustment method is detailed in 3.2 section .