Validate Model Performance
After uploading or converting a custom model, verify it runs correctly on reCamera Pro before deploying it in production.
Step 1: Start inference
- In the Web UI, go to AI Inference → Inference Configuration.
- Enable inference.
- Select your model as the Running Model.
- Set the inference frequency.
- Click Save Configuration.
Step 2: Check real-time monitoring
Go to Real-time Inference Monitoring to observe:
| Metric | What to check |
|---|---|
| Inference Status | Running or stopped |
| FPS | Frames per second — higher is better |
| Detection Results | Class names, confidence scores, bounding boxes |
| Detection Count | Number of objects per frame |
Step 3: Validate detection accuracy
- Point the camera at test subjects matching your detection categories.
- Observe whether detections appear with correct class names and reasonable confidence scores.
- Check for false positives (objects detected that are not present) and false negatives (objects present but not detected).
Adjusting post-processing parameters
If detection quality is poor, revisit Model Configuration:
| Parameter | Effect |
|---|---|
| IOU threshold | Lower → fewer duplicate boxes; higher → more overlapping boxes kept |
| Confidence threshold | Lower → more detections (including false positives); higher → fewer but more reliable detections |
| max_obj | Maximum objects reported per frame |
Step 4: Check resource usage
Monitor device resource usage during inference via SSH:
# CPU and memory
top -bn1 | head -5
# NPU usage (if available)
# TODO(verify): confirm the command to check RKNN NPU utilization on reCamera Pro
Common issues
| Symptom | Possible cause | Fix |
|---|---|---|
| Model fails to load | Incompatible operator or quantization format | Re-convert with RKNN-Toolkit2; check Model Compatibility |
| Very low FPS | Model too large or complex | Use a lighter model or reduce input resolution |
| No detections | Wrong category configuration or threshold too high | Verify detection categories match model output; lower confidence threshold |
| Garbled class names | Category list mismatch | Re-import correct category TXT in Model Configuration |
Related pages
- Choose a Model Deployment Path
- Upload and Configure an RKNN Model
- Configure Detection
- Model Compatibility
Tech Support & Product Discussion
Thank you for choosing our products! We are here to provide you with different support to ensure that your experience with our products is as smooth as possible. We offer several communication channels to cater to different preferences and needs.