Online Training Platform |
1. Training datasets stored in the cloud, providing storage and computing resources;
2. Supports online annotation, allowing data import and training annotation;
3. Supports dataset visualization, model training, testing, and deployment package generation;
4. The deployment package includes executable files that can be directly run on the board, and provides C++ and MicroPython deployment source code. After the task is completed, the deployment package is sent to the registered email; |
1. Only supports image classification and object detection task types;
2. Data security is not sensitive; |
Suitable for users without computing resources, no data security requirements, and those familiar with embedded development; |
1. Linux+RT-Smart dual system image;
2. MicroPython image; |
AI Cube |
1. Training data stored locally, ensuring data security and control;
2. Supports a wide range of tasks, including image classification, object detection, semantic segmentation, OCR detection, OCR recognition, metric learning, multi-label classification, and anomaly detection;
3. Rich autonomous features, supporting basic functions such as model training, testing, and deployment package generation, as well as detailed features like data visualization, dataset viewing, training parameter configuration, data augmentation visualization, and independent data testing;
4. The deployment package includes executable files that can be directly run on the board, and provides C++ and MicroPython deployment source code;
5. Supports both Ubuntu and Windows platforms; |
1. Users need to configure their own computing resources;
2. The installation package is relatively large due to the inclusion of various task training environments; |
Suitable for users with computing resources, familiar with embedded development, and have data security requirements; |
1. Linux+RT-Smart dual system image;
2. MicroPython image; |
k230_training_scripts (KTS) |
1. Provides k230 AI development examples for different modalities such as CV, NLP, and speech;
2. Supports end-to-end comprehensive teaching, covering detailed steps such as environment setup, model training, testing, conversion, image compilation, and on-board debugging;
3. Each step is flexible and controllable, allowing users to replace models, adjust parameters according to their needs, or directly use other pre-trained models to configure calibration sets for model conversion, and modify C++ code for task adaptation; |
1. Code-intensive, requiring strong hands-on skills from users;
2. High resource requirements, needing corresponding server resources or personal computing resources; |
Suitable for users with high data security needs, who enjoy hands-on development, debugging source code, and have certain server and computing resources; |
1. Linux+RT-Smart dual system image;
2. MicroPython image;
3. Pure RT-Smart single system image, the image needs to be compiled by yourself; |
CanCollectorPlus |
1. Serves as a supplement to the above three tools, obtaining datasets collected by k230;
2. The collected data is used as training data, maintaining consistency with data collected by the development board during deployment, reducing differences in color, lighting, angle, resolution, etc. |
1. Users need to complete the annotation themselves; |
Suitable for scenarios where models trained on public datasets perform differently when deployed on the k230 development board; |
1. Linux+RT-Smart dual system image; |