Training sample collection tools

Available with Image Server

Training samples represent a feature of interest or representative pixels and are used to train deep learning models. The first task is to prepare the training samples, followed by exporting the samples for use in creating the deep learning model. Several tools are available in the Prepare training data step to complete the work necessary to export image chips to train a deep learning model.

Tool	Description
Collect training samples	Collect samples using the input imagery layer and basemap for context in individual work units. Use editing tools to create training samples. Rectangle Circle Polygon Select sample Delete Layers Basemap Filter Tip: For the Delete button to be active, a sample must be selected in the viewer. You can add contextual layers to the map to improve the training sample collection process.
Import training samples	Import existing feature layers. Import feature layers from other Deep Learning Studio projects in your organization. Import work units from other Deep Learning Studio projects in your organization.
Edit training sample labels	Edit existing training sample labels. If a schema contains more than one label, you can switch existing sample labels.
Review training samples	Review collected samples and measure the training sample collection progress in a dashboard interface. Review work units individually. View samples in map or table view.
Manage image chips	Export approved training samples as image chips for model training. Register training samples that were created outside of Deep Learning Studio. Maintain records of labeled image chips created in Deep Learning Studio. View exported labeled image chips in the list. View examples of image chips to review the process.
Manage label schema	Add labels to the schema. Import label schemas from other Deep Learning Studio projects. Edit or rename existing labels.
Edit instructions URL	Add a website for instructions on label creation.

Note:

Training samples are labeled according to the label schema, and labels are retained when image chips are created from the training samples.