Skip To Content

Workflows

Available with Image Server

In Deep Learning Studio, you manage the deep learning process using one of two workflows depending on the desired output.

Suggested workflows

Since each deep learning project can have different inputs and different team members, the workflows in Deep Learning Studio are flexible and steps can be customized. The two Deep Learning Studio workflows are a complete workflow and a custom workflow. To determine which workflow is appropriate for a deep learning project, ask yourself the following questions:

  • Which imagery data source will be used?
  • Which training sample schema will be used?
  • How will the collaboration be managed? Which existing groups in ArcGIS Enterprise will be used or will groups be created? (This question is for projects that will have additional group members.)
  • Which deep learning inferencing tool will be used? Which will create the desired output? (You choose the inferencing tool type when you create a project, and the tool cannot be changed.)

Complete workflow

If you require a beginning-to-end solution that guides you through the deep learning process of creating training samples, training the model, and creating output using one of the inferencing tools, consider the complete workflow.

The image below shows the complete workflow for a deep learning project. There may be additional data that can be used in the process to supplement the training sample collection and model training.

Complete workflow for Deep Learning Studio

  • Create project—This step begins the Deep Learning Studio process by creating an item in your organization that organizes deep learning model development and use. The creation of the project is the first step to complete when you begin, and defines the type of inferencing tool to be used.
  • Prepare training samples—This step includes the creation of training samples by selecting and labeling features of interest, qualities of features, or labeling pixels according to the type of deep learning inferencing that will be used for analysis. There are many substeps to this step. When you select it in the app, not all substeps will be necessary, but having a set of approved training samples allows you to train the deep learning model in the next step.
  • Train model—This step includes the creation of the deep learning model by training it with samples based on parameters set in the step. Existing models can be used as the basis for new models, allowing customization for specific features.
  • Use Inferencing tool—This step uses the deep learning model to detect or classify imagery according to the inferencing tool used. For each inferencing tool, there are additional model parameters and options you can use to customize the output.
  • Review results—When the inferencing tool has completed, the result is visible in a map from the app. Evaluate the results of the inferencing process and determine if the output generated is acceptable for the analysis.
  • Share results—This step shares results, but intermediate steps can also be shared. When the deep learning process is completed, the feature layer or imagery layer shows the desired features or labeled pixels. The results can be shared through the app or as an item in your organization.
.

Custom workflow

If some of the steps in the workflow have been completed outside the Deep Learning Studio project, you can use the custom workflow to complete the analysis. Depending on the data and processing that have been completed, different tools may be used in the custom workflow for each project. For example, if the deep learning project does not need a model to be trained, you can use the inferencing tool from the Use Inferencing tool step from the complete workflow in the custom workflow. If a deep learning model does not exist, but the training information for the model has been collected, you can use the Train model step from the complete workflow to train the deep learning model for inferencing in the custom workflow.

Using the custom workflow, you can perform the steps in the order that best suits the project. Steps can be performed iteratively, and you can repeat steps if necessary.


In this topic
  1. Suggested workflows