AI Trainer Operation Guide

1.  **Find the training button in VG3 and click it to open the training interface**

![](/media/202601//1768534233.941474.png)

2.  **After clicking the training interface, click New Project (if you have an existing project, you can click on it; the first time you enter the training interface, you must click the created project)**

![](/media/202601//1768534234.244069.png)

**3. After clicking New Project, the New Project page will appear. On this page, you can customize the project name and project path**

![](/media/202601//1768534234.4543793.png)

4.  **After setting the project name and project path, click OK to enter the dataset interface**

![](/media/202601//1768534234.6766684.png)

5.  **If you need to enter a previously created project, click the project to return to the project page, and then click the old project**
![](/media/202601/pic_5_1768543799.png)

![](/media/202601/pic_5_1_1768543842.png)

6.  **Create a dataset in the data interface. Click the plus sign in the red box to create a new dataset, and the minus sign next to it to delete a dataset**

![](/media/202601//1768534235.4092915.png)

7.  **Click Add Dataset, customize the dataset name, and then import the data**

![](/media/202601/pic_6_1768543871.png)

8.  **After clicking Import Data, find the folder path where the data is located, click Select Folder, and then click OK**

![](/media/202601//1768534235.8663375.png)

![](/media/202601//1768534236.109356.png)

9.  **A successfully created dataset is shown in the figure below**

![](/media/202601//1768534236.3434885.png)

10. **If you need to add data to this newly created dataset, click Fill Data in the marked area as shown below**

![](/media/202601//1768534236.5685623.png)

As shown in the figure below, Marker 1: Display large image
 Marker 2: Display image list
 Marker 4: Delete image
 Marker 5: Check images when deleting

![](/media/202601//1768534236.798829.png)

11. **Import the annotation list. Click Annotate (Marker 1) to enter the annotation page, then add the annotation list (Marker 2)**

![](/media/202601//1768534237.0342202.png)

12. Click the Add button. The Add Task window will pop up. In this window, you can customize the annotation name (Marker 1) (for example, this annotation was modified to 'Cosmetic Cotton Swab Detection'). Under Annotation Type, there are three options: Rectangle Box, Rotated Rectangle Box, and Classification, corresponding to different training models (Marker 2). The choice of annotation type depends on the task requirements. Different annotation type choices will affect subsequent model training operations and also affect the annotation operations here. After selecting the annotation type, check the corresponding dataset to be annotated (Marker 3), and then click OK.

![](/media/202601//1768534237.266574.png)

13. **After clicking OK, you will enter the annotation page, as shown below**

Marker 1: This button can be used to delete and adjust annotation boxes
Marker 2: This button can be used for drawing annotation boxes (rectangular boxes here)
Marker 3: This button can be used for adding fixed rectangle boxes, with the width and height of the fixed rectangle box adjustable nearby
Marker 4: Here you can add required labels, delete labels, and import external labels
Marker 5: This button is used for quick classification annotation
Marker 6: This area displays images in different states
Marker 7: These buttons are for switching images. The corresponding shortcut keys are A or ◀ for switching left, and D or ▶ for switching right.
Marker 8: Displays annotation-related information

![](/media/202601//1768534237.5107222.png)

14. **Add labels before annotating.** Click the plus sign at Marker 1 in the figure below to add a label. In the pop-up Add Type window, add the corresponding type name. If needed, you can click Marker 3 to customize the color, and then click Confirm.

![](/media/202601//1768534237.7581038.png)

The figure below shows a successful label addition.

![](/media/202601//1768534237.987559.png)

If you need to delete a label, first click Marker 1 to select the label to be deleted, then click Marker 2. A prompt as shown in Marker 3 will appear; select OK in Marker 3 to delete.

![](/media/202601//1768534238.2328408.png)

15. **Annotate the data**

Step 1: Select the required label name. As shown at Marker 1 in the figure below, selecting 'Dent' will highlight it in a darker color.

Step 2: Click on Select and Drag to Add, as shown at Marker 2.

Step 3: Drag the mouse to select the area that needs annotation. Press and hold the left mouse button, drag the mouse, and release to complete the annotation, as shown at Marker 3.

Marker 4 shows the completed annotations.

After completing the annotation for one image, switch to the next image and repeat the above steps (you can use shortcut keys A and D to switch images).

![](/media/202601//1768534238.4475017.png)

Fixed box annotation can also be used during annotation.

1.  Select 'Selection Add' in Marker 1.
2.  Choose the appropriate width and height in Marker 2.
3.  Select the area to be annotated and confirm the label (Marker 3).
4.  Marker 4 shows the result after clicking to annotate.

![](/media/202601//1768534238.7110298.png)

16. **After annotating all images, click Training to enter the training interface (as shown in the marked area).**

![](/media/202601//1768534238.9472702.png)

17. In the training list column, add a training list. First, click the plus sign in the training list to open the Add Model window, as shown at Marker 1. Set the relevant information in the Add Model window. The model name can be customized (e.g., changed to 'Cosmetic Cotton Swab Model Training' in the figure below). The model type can be selected according to requirements (here, Object Detection is selected based on the annotation type). Then, check the dataset in the task column and click OK to add the dataset.

![](/media/202601//1768534239.1658502.png)

![](/media/202601//1768534239.3993716.png)

![](/media/202601//1768534239.6425855.png)

18. **Set the relevant training parameters.**

The training/inference device depends on the equipment configuration. If you have an NVIDIA graphics card, you can select GPU. The model scale can be chosen independently. With the same configuration, a larger model scale results in slower training and inference speed, while a smaller model scale results in faster training and inference speed. The scaling factor affects training speed and is generally set based on the size of the training images, and should generally not exceed the length or width of the training image size. Typically, if the image length or width is greater than 1000 pixels, set the scaling factor to 16-20. The scaling factor can also be set according to actual needs. For the original image size, a smaller scaling factor leads to greater precision loss but faster inference speed in VG3. Correspondingly, a larger scaling factor for the original image size leads to higher precision but slower inference speed in VG3. The learning rate can be set as shown in the figure below. The number of epochs can be set according to actual needs. Generally, based on the dataset size (i.e., more images can correspond to a lower number of epochs), you can also increase the number of epochs incrementally based on the inference results after training to achieve the optimal result. The batch size is set according to the GPU memory size. A larger batch size consumes more memory.

![](/media/202601//1768534239.900874.png)

19. **Train.** After setting the parameters, you can click Train.

![](/media/202601//1768534240.133913.png)

20. **Exporting the model.** Right-click the marked area, as shown in the figure below, and click Export Model.

![](/media/202601//1768534240.3577168.png)

21. **After clicking Export Model, a small Export Model window will pop up, as shown below.**

![](/media/202601//1768534240.598797.png)

22. **In the Export Model window, set the model name and the location where the model should be exported. Follow the instructions as prompted in the figure below.**

![](/media/202601//1768534240.8223467.png)

Then click OK to perform the model export operation.

![](/media/202601//1768534241.0658307.png)

The figure below indicates that the model export was successful.

![](/media/202601//1768534241.3232193.png)

23. **Import the model in VG3**

First, find the AI Model Tree on the far right side of VG3 and click it. As shown below.

![](/media/202601//1768534241.544919.png)

After clicking the AI Model Tree, the operation interface for adding an AI model will pop up, as shown below. Click Add.

![](/media/202601//1768534241.769563.png)

After clicking Add, an interface as shown in the figure below will pop up. The model name can be customized. For Thread Selection, select Workflow 2 here. For Algorithm Source, select Zhunyi Technology. The Module Category is selected based on the model type; here, Detection is selected. Finally, click Change Model Path.

![](/media/202601//1768534242.001458.png)

After clicking Change Model Path, locate the model position exported after previous training. Select the folder and click Save.

![](/media/202601//1768534242.2229238.png)

![](/media/202601//1768534242.4578867.png)

Check if the model has been added. It shows the model has been added here.

![](/media/202601//1768534242.689542.png)

Find the Object Detection module under Deep Learning. Double-click or drag and drop to add this module to Workflow 2. In the model selection, choose the model that was just added.

![](/media/202601//1768534242.9799213.png)

![](/media/202601//1768534243.2135608.png)

24. **Training Package Download Location**

[https://www.zhunyitec.com/col.jsp?id\=164](https://www.zhunyitec.com/col.jsp?id=164)

25. **Notes**

All file names and file paths must not contain spaces.

Image file names must also not contain spaces.

Click here to jump to the corresponding Chinese manual page: [AI Trainer Operation Guide](/doc/307)