Start & Monitor Training
Start Training
- Go to the "Training" page, click "New Training Job"
- Configure parameters
- Click "Start Training"
Real-time Monitoring
Once training begins, the detail page displays in real time:
- Loss curves — Box Loss and Class Loss over time
- Accuracy curves — mAP50, mAP50-95, Precision, Recall
- Training progress — current epoch / total epochs
- Estimated time — remaining training time
Actions During Training
- Stop Training: You can interrupt at any time — the trained weights will be saved
- View Logs: Detailed training log output is available
Training Completion
When training finishes:
- The best-performing epoch weights are automatically saved
- You can proceed directly to export
- You can also test the model in your browser first
FAQ
Q: How many images are needed to train a model? A: At least 50 per class is recommended, 200+ for better results. Images should cover different lighting, angles, and backgrounds.
Q: How long does training take? A: It depends on image count and epochs. For 1000 images and 100 epochs with yolov8s, approximately 15-30 minutes.
Q: Can I stop training midway? A: Yes. After clicking "Stop Training", the best-performing weights from completed epochs are saved and can still be exported.
Q: What if Loss stops decreasing? A: Possible causes: ① annotation quality issues; ② learning rate too high/low; ③ insufficient epochs. Check annotation accuracy first, and try increasing epochs.
Q: Why is my job stuck in the queue? A: GPU resources are limited — up to 4 training jobs can run concurrently. You can check queue status on the training page.