Model Input/Output

.. meta:: :description: OpenVINO™ Runtime includes several methods for handling model inputs and outputs.

.. toctree:: :maxdepth: 1 :hidden:

model-input-output/changing-input-shape model-input-output/dynamic-shapes model-input-output/string-tensors

ov::Model::inputs() and ov::Model::outputs() methods retrieve vectors of all input/output ports.

Note that a similar logic is applied to retrieving data using the ov::InferRequest methods.

.. tab-set::

.. tab-item:: Python :sync: py

  .. doxygensnippet:: docs/articles_en/assets/snippets/ov_model_snippets.py
     :language: cpp
     :fragment: [all_inputs_ouputs]

.. tab-item:: C++ :sync: cpp

  .. doxygensnippet:: docs/articles_en/assets/snippets/ov_model_snippets.cpp
     :language: cpp
     :fragment: [all_inputs_ouputs]

ov::Model::input() and ov::Model::output() methods retrieve vectors of specific input/output ports. To select the ports, you may use:

• no arguments, if the model has only one input or output,

• the index of inputs or outputs from the original model framework,

  .. tab-set::

  .. tab-item:: Python :sync: py

  Copy

    .. code-block:: python
  
        ov_model_input = model.input(index)
        ov_model_output = model.output(index)
  

  .. tab-item:: C++ :sync: cpp

  Copy

    .. code-block:: cpp
  
       auto ov_model_input = ov_model->input(index);
       auto ov_model_output = ov_model->output(index);
  

• tensor names of inputs or outputs from the original model framework.

  .. tab-set::

  .. tab-item:: Python :sync: py

  Copy

    .. code-block:: python
  
       ov_model_input = model.input(original_fw_in_tensor_name)
       ov_model_output = model.output(original_fw_out_tensor_name)
  

  .. tab-item:: C++ :sync: cpp

  Copy

    .. code-block:: cpp
  
       auto ov_model_input = ov_model->input(original_fw_in_tensor_name);
       auto ov_model_output = ov_model->output(original_fw_out_tensor_name);
  

Since all ov::Model inputs and outputs are always numbered, using the index is the recommended way. That is because the original frameworks do not necessarily require tensor names, and so, ov::Model may contain an empty list of tensor_names for inputs/outputs. The get_any_name and get_names methods enable you to retrieve one or all tensor names associated with an input/output. If the names are not present, the methods will return empty names.

For information on how ov::InferRequest methods retrieve vectors of input output ports, see the :doc:Inference Request <inference-request> article.

For more details on how to work with model inputs and outputs, see other articles in this category:

• :doc:Changing Input Shapes <model-input-output/changing-input-shape>
• :doc:Dynamic Shape Models <model-input-output/dynamic-shapes>
• :doc:String Tensors <model-input-output/string-tensors>