Edge Developer Toolbox Developer Guide
Create and Train a New Machine Learning Model
On the Edge Developer Toolbox homepage, click on Create AI Model to access curated tools for model training. The Edge Developer Toolbox leverages the Intel® Geti™ platform for model training.
In the Tools section, click Open.
Previously created AI models (if any) are shown under My Workspace -> AI Models. Click Create new model to start a fresh model training experience.
Enter a unique model name, such as “Defect-Detection” and enter one or more labels for this model. The labels are case sensitive and would be named after the item(s) you want to detect. The labels must match the ones used in the dataset and the Annotations file. For this example, the labels are “box”, “shipping label”, and “defect”. Click Create Label after each new label is added. Under Select Objective, click Detection, and click Next to continue.
After a few seconds, the Select Data Set window opens, and at this point, a placeholder is created in the Edge Developer Toolbox file system for the model, and at the same time, a project is also created for it on the Intel® Geti™ platform.
Click Import Data Set and select a dataset from either your local system, cloud services (AWS* S3 Cloud and Azure Cloud* are supported), or the Intel® Tiber™ Edge Platform file system. Currently, only the Common Objects in Context (COCO) dataset format is supported.
For this example, we will use a dataset stored in the local filesystem.
Click Local System and click Next.
If using Cloud Services, provide relevant Azure or AWS S3 credentials to import the dataset.
Provide a Name for the data set, upload image files for training, and select the Annotation file. Click Upload.
Select the dataset and click Next.
After a few seconds, the Select Model Type window opens. The options include ATSS, SSD, YOLOX, and AUTO which is the default. If AUTO is selected, Geti will select the best model for the task. Pick an option and click Train Now to continue.
The training begins and will take a few minutes to complete. Monitor the training progress under the My Workspace -> AI Models section. Click the button on the left of any model name to see expanded details such as Model Type, Accuracy, Status, and Model path.
After the first model training is completed, you would want to have at least one more model for comparison. For example, you can use a different topology such as ATSS or SSD, or a different dataset for comparison. Repeat steps 1 to 11 to train more models as needed.
After the models are trained, you can proceed to the next step to assess the models’ performance.