Yolov5 realizes road crack detection

Yolov5 Realize road crack detection

GitHub Reference code link

I modified the code Baidu network disk link

password ：2mzl

Dataset Links

password ：06dj

be based on Pytorch Of Yolov5 Operation instructions for road crack detection program . You can use it in combination with my description and the original description , Any questions are welcome .

List of articles

• Yolov5 Realize road crack detection
• • Environmental requirements
  • test
  • Train your own dataset
  • • 1. establish dataset.yaml file
    • 2. Create label file
    • 3. Organization file path
    • 4. Choose a model
    • 5. Start training

Environmental requirements

Python 3.8 Or later

And install requirements.txt All dependent packages in the file , Include 1.7 And above torch

$ pip install -r requirements.txt

test

Put pictures or video files in and before running detect.py In the same directory , Then run the following statement ：

$ python detect.py --source 20200827153531.mp4  # video
							file.jpg  # image 

Because I have put the trained model into ./runs/train/exp_1000/weights/ The path is down , If you train your model , Remember to change to your own model path .

Original drawing mark ： Test mark ：

Because you can't play video , So it was uploaded to Baidu cloud disk , You can take your own （ Contains the original video and results ）.

https://pan.baidu.com/s/1DEu0TYcdowtt6k_A1wOxTQ
 password ：2mzl

Train your own dataset

1. establish dataset.yaml file

The file shall meet the following format （ Here's the picture ）：

1. Download address （ Don't worry about what you don't have ）
2. Training picture path
3. Verify image path
4. Number of classes
5. Class name

# download command/URL (optional)
download: https://github.com/ultralytics/yolov5/releases/download/v1.0/coco128.zip

# train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/]
train: ../coco128/images/train2017/
val: ../coco128/images/train2017/

# number of classes
nc: 80

# class names
names: ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus', 'train', 'truck', 'boat', 'traffic light',
        'fire hydrant', 'stop sign', 'parking meter', 'bench', 'bird', 'cat', 'dog', 'horse', 'sheep', 'cow',
        'elephant', 'bear', 'zebra', 'giraffe', 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee',
        'skis', 'snowboard', 'sports ball', 'kite', 'baseball bat', 'baseball glove', 'skateboard', 'surfboard',
        'tennis racket', 'bottle', 'wine glass', 'cup', 'fork', 'knife', 'spoon', 'bowl', 'banana', 'apple',
        'sandwich', 'orange', 'broccoli', 'carrot', 'hot dog', 'pizza', 'donut', 'cake', 'chair', 'couch',
        'potted plant', 'bed', 'dining table', 'toilet', 'tv', 'laptop', 'mouse', 'remote', 'keyboard', 
        'cell phone', 'microwave', 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock', 'vase', 'scissors', 
        'teddy bear', 'hair drier', 'toothbrush']

2. Create label file

The file shall meet the following format （ Here's the picture ）：

1. One picture one txt file
2. One goal per line
3. Every line is class x_center y_center width height The format of , That is, the sequence number corresponding to the class , Target x Axis center point , Target y Axis center point , And width and height , Pay attention to no more than 1, Are pixels divided by width or height .（ Normal data sets generally have their own labels in this format , If there is no such format , You need to write your own program to convert , If there are no labels, only pictures , You need to download the marking software by yourself , Then mark the picture ）
4. The serial number from 0 Start

0 0.9583333333333334 0.9408333333333334 0.07333333333333333 0.08833333333333333
2 0.7958333333333334 0.8391666666666667 0.4083333333333334 0.04833333333333334
0 0.4083333333333334 0.8508333333333334 0.17666666666666667 0.12166666666666667

3. Organization file path

Pay attention to the 1 Step establish dataset.yaml file File path in , Put the file in the path you fill in .

4. Choose a model

Recommended choice YOLOv5s, Small and fast .

5. Start training

If I use theta Pycharm Right click train.py file open in terminal, Enter the following code , If not Pycharm, You can do it again cmd in , Transfer to train.py Under the path , Then run the following statement to start training .（-- After that, it represents the parameter ,img Is the size of the image to be scaled , It is better to be the same size as the original drawing ,epochs Is the number of iterations ,data Is the file created in the first step ,weight That is, the trained weight ）

# Train YOLOv5s on COCO128 for 5 epochs
python train.py --img 640 --batch 16 --epochs 5 --data coco128.yaml --weights yolov5s.pt
CO128 for 5 epochs
python train.py --img 640 --batch 16 --epochs 5 --data coco128.yaml --weights yolov5s.pt

If there is a problem , You can add the following QR code for consultation