Model Input/Output#

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.

Python

    inputs = model.inputs
    outputs = model.outputs

C++

/* Take information about all topology inputs */
auto inputs = model->inputs();
/* Take information about all topology outputs */
auto outputs = model->outputs();

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,
Python
ov_model_input = model.input(index) ov_model_output = model.output(index)
C++
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.
Python
ov_model_input = model.input(original_fw_in_tensor_name) ov_model_output = model.output(original_fw_out_tensor_name)
C++
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 Inference Request article.

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