Available with Image Server
The Train model step creates a deep learning model that can be used for deep learning analysis throughout ArcGIS. Several parameters are available to control the creation of the deep learning model.
After the labeled image chips are created and saved from the training samples, the next step is to train the deep learning model for inferencing. In the Train model step, you can specify the training iteration, choose a pretrained model, or configure the parameters for training the model. The configuration parameters used to train the model are described in the following table:
| Parameter | Description |
|---|---|
| Pretrained model | Use an existing model. Select the model to be used for training. |
| Choose model type | Choose the model type that will be used when training the model from the drop-down list. |
| Max epochs | Set the number of times that the learning algorithm will work through the entire training dataset. |
| Batch size | Set the number of samples that will be processed through the network. |
| Learning rate | Specify the rate at which new information will be acquired through the training process. |
| Backbone model | Specify which preconfigured deep learning neural network will be used to create the new model. |
| Percentage of images for validation | Specify the portion of training samples to use for model validation. |
| Finish training when model stops improving | Stop the model training when improvement stops. |
| Freeze model | Prevent the weights and biases of the backbone layers in the pretrained model from being modified. |
Note:
All of these parameters are related to the deep learning process. For more information about the parameters, review Deep learning in Raster Analysis.
After configuring the model, you can use it for inferencing or modify the options and rerun the model until the desired results are achieved. Once the model is trained, the next step is to use the model for inferencing. Choose the Run inference step to complete the deep learning process.