Model Training Platform
A cloud-based AI vision model training platform. No coding required — upload data → annotate → train → deploy, all in one place for custom YOLO object detection model training and deployment.
Platform URL: https://aihub.shimetapi.cn/
What is the Model Training Platform?
The Model Training Platform is a zero-code AI vision model customization platform. Simply upload images, draw bounding boxes around target objects, and the platform automatically trains a YOLO object detection model on cloud GPUs. Once training is complete, export the model in a format ready to run on your edge device.
Typical Application Scenarios
| Scenario | Examples |
|---|---|
| Safety Monitoring | Hard hat detection, fire/smoke detection, intrusion detection |
| Smart Kitchen | Chef hat detection, mask detection, rodent detection |
| Industrial Production | Defect detection, foreign object detection on assembly lines |
| Smart Transportation | Vehicle detection, license plate recognition, people counting |
| Behavior Recognition | Fall detection, specific action recognition |
Platform Advantages
- Zero-code: Entirely web-based — no Python or deep learning knowledge required
- Cloud GPU: 4× NVIDIA RTX 6000 Ada 48GB professional GPUs for fast training
- Real-time monitoring: Live Loss curves and accuracy curves during training
- End-to-end closed loop: From data upload to edge device deployment, one platform does it all
- Team collaboration: Multi-user simultaneous annotation, shared datasets, admin/member role control
- Multi-platform export: Supports RKNN (RK3588), XMM (GK7206), and other mainstream edge NPU formats
Core Workflow
The platform follows a complete upload → annotate → train → export → deploy closed loop:
┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
│ ① Upload │ → │② Annotate │ → │ ③ Train │ → │ ④ Export │ → │ ⑤ Deploy │
│ Dataset │ │ BBoxes │ │ GPU │ │ RKNN/XMM │ │ Edge │
└──────────┘ └──────────┘ └──────────┘ └──────────┘ └──────────┘| Stage | What You Do | What the Platform Does |
|---|---|---|
| Upload | Create a dataset, upload images (batch upload or real-time phone capture supported) | Auto-process images and sync to training server |
| Annotate | Draw bounding boxes on target objects, assign class labels to each box | Real-time save of annotation data, annotation progress tracking |
| Train | Select a base model, set epochs, batch size, and other parameters, then click start | GPU server auto-trains, real-time Loss/mAP metrics pushed |
| Export | Choose target platform and export format, click export | Auto model conversion (PT→ONNX→RKNN/XMM) with automatic quantization |
| Deploy | Download the model file and deploy to your edge device | Provides model download link |