Deploy and Train General Models

Makeblock Customer Service Team

September 16, 2025 03:24

AI Camera 2.0 supports smart recognition using custom models.

1. Preparations

AI Camera 2.0
mTraining web platform

2. Train custom models

Step 1: Create a training project

Log in to the mTraining web platform.
Choose Training Projects > New Project.

Set the project name, select the task type, and click Start Training.
- Image Detection (identify object locations)
- Image Classification (identify object categories)

Step 2: Build a dataset

Open the project you just created and choose Select Dataset > Create New Dataset.

Specify Dataset Name and select Dataset Type (must match the project type), and then click Create Dataset.

Prepare images using one of the following methods:

Method 1: Upload from your computer – click Upload Images to import local files.

Method 2: Capture images with the device:

a. Click Device Collection.

b. In the Device Collection section on the left, select Training Set and click Generate QR Code.

c. On AI Camera 2.0, tap Self-trained Model, or use the corresponding coding block to start this feature.

d. Click Collect Image and scan the QR code on mTraining to open the photo capture interface. (Note: QR codes are valid for 5 minutes; refresh if expired.)

e. Aim AI Camera 2.0 at the target and use the Photo function. When the text Uploaded is displayed in the lower left corner, the image has been successfully uploaded.

Step 3: Annotate data

In the dataset page, click Annotate Data to enter the image annotation interface.

Click Add Label and create the required categories (e.g., Dog, Cat, Car).

Annotate each image:
- Object Detection: Draw a box around the target, select the label, and click Save Annotation.
- Image Classification: Assign a label to the entire image.

Once all images are labeled and categorized, the dataset annotation is complete.

Step 4: Train the model

Click Training Projects, bind the annotated dataset to your project, and click Confirm.

The system will display an overview of the dataset.

Click Create Training Task and set the following parameters:
- Name: training task name
- Batch Size: recommended 8–16
- Epochs: typically 50–200
- Maximum Learning Rate: 0.001 by default

Click Create Task to start training after all parameters are set.

Step 5: Deploy the model

Select the trained model and click Deploy to generate a QR code.

2. On AI Camera 2.0, tap Self-trained Model and click Deploy Model.

Scan the QR code to automatically download the trained model.

3. Use and manage models

Use models

Go to the model detection page to recognize learned objects.

Manage models

In My Models view, you can view all models and click Edit to deploy or delete a model.

Coding block example: Cat & dog detection with distance alert

This example uses AI Camera 2.0 to recognize cats and dogs. By combining coordinate analysis with light alerts, this example demonstrates the intelligent perception process from object detection and distance analysis to status feedback.