Model Evaluation & Testing
Browser Inference Test
After training completes, you can test the model directly in your browser without deploying to hardware.
- Go to the "Model Management" page
- Find your trained model and click "Test"
- Upload a test image
- The platform returns detection results: the image with overlaid bounding boxes and class labels
- View confidence scores for each detection
This is suitable for quick model validation. For actual edge inference speed, rely on post-deployment measurements.
Mobile Inference Test
Access the platform from your phone's browser to test models using your phone camera — take a photo and get real-time results.
Performance Expectations
| Model | Training Time (1000 images, 100 epochs) | Typical mAP50 |
|---|---|---|
| yolov8n | ~10 min | 0.45-0.65 |
| yolov8s | ~20 min | 0.55-0.75 |
| yolov8m | ~35 min | 0.60-0.80 |
| yolov8l | ~50 min | 0.65-0.82 |
Supported Model Specs
| Dimension | Range |
|---|---|
| Algorithm | Select Ultralytics YOLO series |
| Task Type | Object Detection |
| Input Size | 640×640 (default, adjustable) |
| Export Formats | PT, ONNX, RKNN, XMM |
Target Deployment Platforms
| Platform | NPU Chip | Export Format | Typical Device |
|---|---|---|---|
| RK3588 | Rockchip RK3588 | RKNN | Edge AI box, AI camera |
| GK7206 | Goke GK7206 | XMM | IPC camera SoC |