Create input for a model

A model requires input tensors with certain shapes and data types. Use this information to find model inputs and create input tensors for your model.

Understand the required input

Before you can create input tensors for a model, you must first understand the shape and data types of the model inputs. For more information, refer to Model inputs.

The TensorShape of the [Tensor](xref:Unity.Sentis.Tensor) you create must be compatible with the SymbolicTensorShape, which defines the shape of the model input.

Convert an array to a tensor

To create a tensor from a one-dimensional data array, follow these steps:

  1. Create a TensorShape object that has the length of each axis.
  2. Create a Tensor object with the TensorShape object and the data array.

For example, the following code creates a tensor for a model that takes an input tensor of shape 3 × 1 × 3.

using UnityEngine;
using Unity.Sentis;

public class ConvertArrayToTensor : MonoBehaviour
{
    void Start()
    {
        // Create a data array with 9 values
        float[] data = new float[] { 1f, 2f, 3f, 4f, 5f, 6f, 7f, 8f, 9f };

        // Create a 3D tensor shape with size 3 × 1 × 3
        TensorShape shape = new TensorShape(3, 1, 3);

        // Create a new tensor from the array
        TensorFloat tensor = new TensorFloat(shape, data);
    }
}

Create multiple inputs

If a model needs multiple input tensors, create a dictionary that contains the inputs.

For example:

Dictionary<string, Tensor> inputTensors = new Dictionary<string, Tensor>()
{
    { "x", xTensor },
    { "y", yTensor },
};

To avoid handling a dictionary, set inputs manually with SetInput.

For example:

worker.SetInput("x", xTensor);
worker.SetInput("y", yTensor);

Edit a model

Use the functional API to add operations to your model inputs. For more information, refer to Edit a model.

Additional resources