To create a new project configuration, import and select a model. On the Select Model page, you can import a model, import one from the Model Zoo , add a path to a local model, or select one from a list of already imported models.

To use a model that has already been imported, select the model in the list and click the Use Selected Model button. Now you can proceed to select a dataset.

If you want to create a project configuration with a new model, you need to import the model.

NOTE: A model must be previously converted to an IR using the Model Optimizer. For details, go to the Model Optimizer Developer Guide.

Import Model

To upload a model, click the Import Model button on the Select Model page and fill the Import Model form:


Model File	A `.zip` or `.tar.gz` archive that contains the model Intermediate Representation (IR)(`.xml` and `.bin` file)
Local Path	A path to a directory on your disk that contains the model IR. Not needed in this step.
Model Name	A unique name for your model. If nothing is entered, this defaults to the archive filename.
Model Type	Can be Classification , Object Detection or Generic.

NOTE: The Advanced Configuration fields are prepopulated with parameters extracted from the Intermediate Representaton .xml. Modifying these parameters impact accuracy calculations and model calibration. For more information, see Configure Accuracy Settings.

After you have entered all of the required data, click the Import Model IR button to start the import. The process of running an initial inference begins and the Status column in the Imported Model table shows the import progress in the progress bar.

Select the model in the list and click the Use Selected Model button. Now you can proceed to select a dataset.

Import from Model Zoo

Another option is to import a model from the Model Zoo, an open source repository of pretrained models.

NOTE: Precision of models from the Model Zoo cannot be changed. The default value is FP32.

NOTE: Internet connection is required for this option.

To import a local model, click the Import Model button on the Select Model page:

Select a model by clicking on it, then press Import.

The process starts and the Status column in the Imported Model table shows the progress bar and the status of the import.

Select the model in the list and click the Use Selected Model button. Now you can proceed to select a dataset.

Add Local Model

Importing and performing inference on large models and dataset samples can take significant time to complete. To import models and datasets from a local file system, you can mount a data directory with the Workbench Docker* container. Refer to Mount Folder to Docker Container for more details.

To import local datasets and models, move a folder with your IR models to the ~/workbench/user_data path.

NOTE: If you mounted your local folder to the container, use the path that you set for it.

On the Select Model page, add the local path to the model.

NOTE: Do not include file names in the directory containing the IR.


Model File	A `.zip` or `.tar.gz` archive that contains the model IR (`.xml` and `.bin` files). Not needed in this step.
Local Path	A path to a directory on your disk that contains the model IR.
Model Name	Can be different from the archive name.
Task	Can be Classification, Object Detection or Generic.

Click on the Add local button. The process starts and the Status column in the Imported Model table shows the progress bar and the status of the import. To remove an imported model from the list, click on the red X icon in the Action column.

NOTE: A model that was added locally is removed only from the DL Workbench page but not deleted. The model remains on the file system.

Select the model in the list and click the Use Selected Model button. Now you can proceed to select a dataset.