Convert model with Model Optimizer


Model Optimizer is a cross-platform command-line tool that facilitates the transition between training and deployment environments, performs static model analysis, and adjusts deep learning models for optimal execution on end-point target devices.

Using Model Optimizer tool assumes you already have a deep learning model trained using one of the supported frameworks: TensorFlow, PyTorch, PaddlePaddle, MXNet, Caffe, Kaldi, or represented in ONNX* format. Model Optimizer produces an Intermediate Representation (IR) of the model, which can be inferred with OpenVINO™ Runtime.


Model Optimizer does not infer models. Model Optimizer is an offline tool that converts a model into IR and optimizes before the inference takes place.

The scheme below illustrates the typical workflow for deploying a trained deep learning model:


The IR is a pair of files describing the model:

  • .xml - Describes the network topology

  • .bin - Contains the weights and biases binary data.


The generated IR can be additionally optimized for inference by Post-training optimization that applies post-training quantization methods.


You also can work with the Model Optimizer inside the OpenVINO™ Deep Learning Workbench (DL Workbench). DL Workbench is a web-based graphical environment that enables you to optimize, fine-tune, analyze, visualize, and compare performance of deep learning models.

Run Model Optimizer

To convert the model to IR, run Model Optimizer:

mo --input_model INPUT_MODEL

If out-of-the-box conversion (only the --input_model parameter is specified) is not succeed, try to use parameters for overriding input shapes and cutting the model, mentioned below.

To override original input shapes for model conversion, Model Optimizer provides two parameters: --input and --input_shape. For more information about these parameters, refer to Setting Input Shapes.

To cut off unwanted parts of a model, such as unsupported operations and training sub-graphs, the --input and --output parameters can be used, defining new inputs and outputs of the converted model. For a more detailed description, refer to Cutting Off Parts of a Model.

Also, you can insert additional input pre-processing sub-graphs into the converted model using the --mean_values, scales_values, --layout, and other parameters described in Embedding Preprocessing Computation.

Model Optimizer’s compression parameter --data_type allows to generate IR of the FP16 data type. For more details, please refer to Compression of a Model to FP16.

To get the full list of conversion parameters available in Model Optimizer, run the following command:

mo --help

Examples of CLI Commands

Below is a list of separate examples for different frameworks and Model Optimizer parameters.

  1. Launch Model Optimizer for a TensorFlow MobileNet model in the binary protobuf format.

    mo --input_model MobileNet.pb

    Launch Model Optimizer for a TensorFlow BERT model in the SavedModel format, with three inputs. Explicitly specify input shapes where the batch size and the sequence length equal 2 and 30 respectively.

    mo --saved_model_dir BERT --input mask,word_ids,type_ids --input_shape [2,30],[2,30],[2,30]

    For more information on TensorFlow model conversion, refer to Converting a TensorFlow Model.

  2. Launch Model Optimizer for an ONNX OCR model and explicitly specify new output.

    mo --input_model ocr.onnx --output probabilities

    For more information on ONNX model conversion, please refer to Converting an ONNX Model. Note that PyTorch models must be exported to the ONNX format before its conversion into IR. More details can be found in Converting a PyTorch Model.

  3. Launch Model Optimizer for a PaddlePaddle UNet model and apply mean-scale normalization to the input.

    mo --input_model unet.pdmodel --mean_values [123,117,104] --scale 255

    For more information on PaddlePaddle model conversion, please refer to Converting a PaddlePaddle Model.

  4. Launch Model Optimizer for an MXNet SSD Inception V3 model and specify first-channel layout for the input.

    mo --input_model ssd_inception_v3-0000.params --layout NCHW

    For more information on MXNet models conversion, please refer to Converting an MXNet Model.

  5. Launch Model Optimizer for a Caffe AlexNet model with input channels in the RGB format, which needs to be reversed.

    mo --input_model alexnet.caffemodel --reverse_input_channels

    For more information on Caffe model conversion, please refer to Converting a Caffe Model.

  6. Launch Model Optimizer for a Kaldi LibriSpeech nnet2 model.

    mo --input_model librispeech_nnet2.mdl --input_shape [1,140]

    For more information on Kaldi model conversion, refer to Converting a Kaldi Model.

To get conversion recipes for specific TensorFlow, ONNX, PyTorch, MXNet, and Kaldi models, refer to Model Conversion Tutorials.