get_trt_engine_inputs_and_outputs

inference_models.models.common.trt.get_trt_engine_inputs_and_outputs

get_trt_engine_inputs_and_outputs(engine)

Extract input and output tensor names from a TensorRT engine.

Inspects a TensorRT engine to determine which tensors are inputs and which are outputs. This is useful for setting up inference execution contexts and understanding the engine's interface.

Parameters:

• engine

  (ICudaEngine) –

  TensorRT CUDA engine (ICudaEngine) to inspect.

Returns:

• Tuple[List[str], List[str]] –

  Tuple of (input_names, output_names) where: - input_names: List of input tensor names - output_names: List of output tensor names

Examples:

Inspect TensorRT engine:

>>> from inference_models.developer_tools import (
...     load_trt_model,
...     get_trt_engine_inputs_and_outputs
... )
>>>
>>> engine = load_trt_model("model.plan")
>>> inputs, outputs = get_trt_engine_inputs_and_outputs(engine)
>>>
>>> print(f"Inputs: {inputs}")   # ['images']
>>> print(f"Outputs: {outputs}") # ['output0', 'output1']

Use for setting up inference:

>>> inputs, outputs = get_trt_engine_inputs_and_outputs(engine)
>>> context = engine.create_execution_context()
>>>
>>> # Set input tensor
>>> input_name = inputs[0]
>>> context.set_input_shape(input_name, (1, 3, 640, 640))

Note

• Requires TensorRT to be installed
• Works with TensorRT 10.x engines
• Tensor names are defined during engine building/export

See Also

• load_trt_model(): Load TensorRT engine from file
• infer_from_trt_engine(): Run inference with TensorRT engine