Create Your First Model (30-Minute Quick Experience)

Step 1: Create a Dataset (2 min)

  1. Click sidebar "Datasets" → "New Dataset"
  2. Enter a name (e.g., "Helmet Detection") and description
  3. Click Create

Step 2: Upload Images (3 min)

  1. Enter the dataset detail page
  2. Drag and drop image folders to the upload area (JPG/PNG supported, max 100MB each) / scan QR code for real-time phone upload
  3. Wait for upload to complete (progress bar shows real-time status)

Step 3: Annotate Images (15 min)

  1. Click any image to enter the annotation editor
  2. Add class labels in the left panel (e.g., helmet, head, person)
  3. Draw bounding boxes by dragging on the image, select the corresponding class
  4. Click "Next" after annotating each image
  5. Recommended: at least 50 images per class

Annotation Editor

AreaContent
Left PanelClass label management, annotation list
CanvasImage + interactive bounding boxes
Top ToolbarZoom, undo, save, page navigation

Annotation Tips

  • Draw tight boxes that closely fit the target object — avoid excessive blank space
  • Annotate the visible portion of occluded objects as well
  • At least 50 instances per class, 200+ recommended for best results
  • Cover different lighting, angles, and background variations
  • Always click "Save" after completing annotations

Step 4: Create a Version Snapshot (1 min)

  1. Return to the dataset detail page
  2. Click "Create Snapshot", enter a version description (e.g., "v1.0 initial version")
  3. The platform auto-generates YOLO-format training data

Step 5: Start Training (1 min setup + 10-30 min wait)

  1. Click sidebar "Training" → "New Training Job"
  2. Select the dataset and snapshot version you just created
  3. Choose a base model (recommended: yolov8s.pt for balanced speed and accuracy)
  4. Set epochs (recommended: 100-300) and batch size (recommended: 16)
  5. Click "Start Training"
  6. Monitor real-time Loss curves and accuracy changes on the training detail page

Base Model Selection

ModelParamsSpeedAccuracyBest For
yolov8n.pt3.2MFastest★★Extreme real-time needs, few classes
yolov8s.pt11.2MFast★★★Balanced choice
yolov8m.pt25.9MMedium★★★★Higher accuracy required
yolov8l.pt43.7MSlow★★★★★High-precision scenarios

Step 6: Export the Model (2 min)

  1. After training completes, click "Deploy Model" — the model will be saved to the model list, then click "Export Model"
  2. Select the target format (RKNN or XMM)
  3. Wait for conversion to complete, then click Download

Next