Detailed explanation of yolov5 training own data set

Beginner little rookie , I hope it's like taking notes and recording what I've learned , Also hope to help the same entry-level people , I hope the big guys can help correct it ~ Tort made delete .

Catalog

One 、YOLOv5 Source download

Two 、 Environment configuration

3、 ... and 、 Create a dataset

Four 、 Change configuration

1、coco128.yaml

2、train.py

5、 ... and 、 Run

6、 ... and 、 Result presentation

---

One 、YOLOv5 Source download

Website Guide ：GitHub - ultralytics/yolov5: YOLOv5 in PyTorch > ONNX > CoreML > TFLite

use git Cloned

---

Two 、 Environment configuration

There is a file in the folder requirements.txt, Here is the description of environment dependency

We input... At the terminal pip install -r requirements.txt Download the installation dependency package

---

3、 ... and 、 Create a dataset

Create your own data set according to your needs

Here you can see previous blogs

Make your own dataset _tt Ya's blog -CSDN Blog

stay YOLOv5 Create a folder in the directory data1 To load our data

then data1 Next is images and labels

notes ： The folder name here should be images and labels, Otherwise, there will be all kinds of errors , The reason why I'm too lazy to go anywhere , Let him run first （label You will also report mistakes. , Add a s Then he behaved ）

---

Four 、 Change configuration

1、coco128.yaml

His default training set is coco128, So I want to be lazy , Directly in coco128.yaml Change it into mine

 coco128.yaml The file in yolov5/data Folder

Here is my configuration

path: ../data1  # dataset root dir
train: images/train  # train images (relative to 'path') 128 images
val: images/train  # val images (relative to 'path') 128 images
test:  # test images (optional)

# Classes
nc: 4  # number of classes
names: ['move','point','up','down']  # class names

 path： Compared with in yolov5 Under the directory , The root of your dataset folder

Move out this picture again

  my data1 stay yolov5 In the last directory of , So it is ../data1

train Is the training data set path ,val Is the path of the validated dataset

Here I follow the original coco128.yaml, These two are the same

So relative to path It is images/train( my images There is also one in the folder train Folder , Then there is my picture ）

nc： Is how many classes you want to detect

names Is the name of these classes

2、train.py

Generally, we only need to change these

 weights It's the model you choose

data： Because I'm using coco128.yaml In itself , So I don't have to change ~

epochs and batch-size Do as you see fit

epochs It is how many times the whole data set will be iterated in the training process , If the graphics card doesn't work, turn it down

batch-size： How many pictures do you see at a time before you update the weight , If the same graphics card doesn't work, turn it down

---

5、 ... and 、 Run

In the terminal cd To yolov5 Directory （ because train.py Under the directory ）

And then directly python train.py That's it

Then wait ~ A long wait ~

---

6、 ... and 、 Result presentation

After the final training, we will yolov5 Create one in the directory runs Folder

There are all kinds of results , I won't say much more

Just say one. weights In a folder best.pt and last.pt

These are weight files , It is the model preservation after training , Can be directly in detect.py Used in documents

Also input... At the terminal

python detect.py --source 0 --weights runs/train/exp/weights/best.pt

there source 0 It refers to the computer camera as the source , Then you can see the training results most intuitively ~

---

You are welcome to criticize and correct in the comment area , thank you ~