Models & Deployment
Models Tab
Q: The Models Tab fails to load
| Possible cause | Solution |
|---|---|
| PC cannot reach aihub.shimetapi.cn | Check the network |
| AIHub token expired | Log in again |
| Server maintenance | Wait or contact official support |
Q: I can't find the model I trained
| Possible cause | Solution |
|---|---|
| Training not finished | The AIHub training job status should be "completed" |
| Not exported to .xmm | After training you must export in XMM format |
| Wrong Tab | Switch to the "User models" sub-Tab |
Q: The model card only shows PT/ONNX
That means XMM format was not selected when exporting after training. Export the XMM again from the training detail page on AIHub.
See the AIHub docs → Export formats.
Q: Can I download models outside the IDE?
Yes. You can also download .xmm.zip from the AIHub web portal, then:
- Workspace → Local models → Import local model
- Select the downloaded zip
Q: How big is a model zip?
| Model | Typical size |
|---|---|
| yolov8n | 6-12 MB |
| yolov5n | 5-10 MB |
| Custom dataset | 5-200 MB |
One-click deploy
Q: Deployment is stuck at "pushing model"
| Possible cause | Solution |
|---|---|
| Board offline | Check the connection status in DEVICE TREE |
| Wrong SSH credentials | The shimetad user can't connect; check firmware |
| Firewall blocks scp | Temporarily disable the firewall |
Q: Deployment succeeded but nothing shows in the Vision Tab
| Possible cause | Solution |
|---|---|
| Model .xmm file corrupted | Deploy again |
| Deployed to the wrong directory | It should be under /mnt/models/xmm/ |
| Vision Tab not refreshed | Switch Tabs and switch back |
Q: Can I deploy multiple models at once?
shimeta-camera (legacy) and shimeta-xmm share port 8080 and cannot run simultaneously. For multiple models:
| Method | Action |
|---|---|
| Run in sequence | Stop A, then start B |
| Multiple boards | Run different models on different Pico-G1s |
Q: Can I view deployment history?
The deployment history UI is not standardized in v1.0.0. You can SSH into the board and check the /mnt/models/xmm/ directory to confirm deployment status.
Choosing a model
Q: How do I choose a model?
| Scenario | Recommendation |
|---|---|
| General object detection | yolov8n |
| Speed first | yolov5n |
| Specific scenario | Search community/official models for the right classes |
Q: How do I know which classes a model supports?
- Official models: shown on the model card
- Community models: see the model detail page
- Self-trained models: set during training on AIHub
Q: Can I download other people's models?
All users can download community models. Click the download button on the model card.
Model market
Q: Official vs community models?
| Type | Source | Who can download |
|---|---|---|
| Official models | Published by the company | All users |
| Community models | Published by users | All users |
Different sources, but all of them can be deployed and run on a Pico-G1.
Q: How do I publish my own model?
Publish from the AIHub web portal. The Workbench IDE does not support publishing directly.
Training models
Q: Can I train inside Workbench?
No. Training is done on the AIHub platform:
- Click the "Train model" link at the top-right of the Models Tab
- The browser opens the training page at aihub.shimetapi.cn
- Finish training + export the .xmm
See the AIHub docs → Training guide.
Q: How are training credits charged?
Training credits are managed on the AIHub platform; the IDE shows no credit UI.
Q: Can I migrate training jobs to the IDE?
No. AIHub training jobs can only be managed in the AIHub web portal.
Model versions
Q: A model has several versions — which one?
Usually pick the latest version (shown by default). Special cases:
| Scenario | Recommended version |
|---|---|
| Maximum accuracy | Latest |
| Maximum speed | Older version (smaller model) |
| Reproducing past experiments | The version used then |
Q: Can I delete old deployed models?
On the board:
ssh shimetad@192.168.49.10
rm -rf /mnt/models/xmm/<model>/Offline scenarios
Q: Can Workbench work fully offline?
Yes, with limited features:
| Feature | Available offline |
|---|---|
| Local model deployment | ✅ Yes |
| Terminal SSH | ✅ Yes |
| Browsing the Models Tab | ❌ No |
| AIHub training | ❌ No |
Q: Can I import a model a colleague gave me?
Yes. Workspace → Local models → Import → select the .xmm.zip.
Quick troubleshooting
| Symptom | Solution |
|---|---|
| Models Tab loads slowly | Check the network; the first load fetches the list |
| Can't find my trained model | Training was not exported to XMM |
| "Deploy to board" unresponsive | Board not connected |
| Deployment fails | Board disk full; clean up old models |