This tutorial shows how to import the SSD MobileNet v2 COCO, one of the original TensoFlow* models, into the DL Workbench. Download the file with the model and unarchive it.
Go to the Import Model page as described in Import Models. Select TensorFlow in the drop-down list. Do not uncheck the Is Frozen Model box, because this is a frozen model. Choose the
frozen_inference_graph.pb file. The name of the model is automatically filled in based on the name of the file you choose first.
Once you click Import Model, the tool analyzes your model and opens the Convert Model to IR form with conversion settings fields. Some of the fields are prepopulated as a result of the analysis.
Choose the BGR color space order as stated in the documentation of the model.
Choose the pipeline configuration file
Choose one of the model conversion configuration files in the drop-down list.
Check the boxes Use Means and Use Scales. Fill in the values as described in the Model Optimizer documentation:
[127.5, 127.5, 127.5] and
[127.5, 127.5, 127.5] respectively.
Once you have the fields of the Convert Model to IR window filled as shown below, click Convert and proceed to select a dataset.