Use a command buffer

You can use a command buffer to create a queue of Sentis commands, then run the commands on the GPU later.

Follow these steps:

  1. Create or get an empty command buffer.
  2. Create a worker with the BackendType.GPUCommandBuffer back end type.
  3. In a Unity event function, for example OnRenderImage, add Sentis methods to the command buffer.
  4. Use Graphics.ExecuteCommandBuffer to execute the command buffer.

To add commands to the command buffer, do any of the following:

  • Use a Sentis API that takes the command buffer as a parameter, for example TextureConverter.ToTensor.
  • Use the ExecuteWorker method on the command buffer to add a command that runs the model.

For example, the following code creates and execute a command buffer queue that converts a render texture to an input tensor, runs the model, then converts the output tensor to a render texture.

using UnityEngine;
using Unity.Sentis;
using UnityEngine.Rendering;

public class UseCommandBuffer : MonoBehaviour
{
    public ModelAsset modelAsset;
    IWorker worker;

    void Start()
    {
        // Create a worker that uses the GPUCommandBuffer back end type
        worker = WorkerFactory.CreateWorker(BackendType.GPUCommandBuffer, ModelLoader.Load(modelAsset));
    }

    void OnRenderImage(RenderTexture source, RenderTexture destination)
    {
        // Create a new command buffer
        CommandBuffer commandBuffer = new CommandBuffer();

        // Add a command to the command buffer that creates an input tensor from the source texture data
        // you can also call TextureConverter.ToTensor(commandBuffer, source, inputTensor) if you allready have the tensor allocated
        Tensor inputTensor = TextureConverter.ToTensor(commandBuffer, source);

        // Add a command to the command buffer that runs the model using the input tensor
        commandBuffer.ExecuteWorker(worker, inputTensor);

        // Add a command to the command buffer that blits the output to the destination render texture
        TextureConverter.RenderToTexture(commandBuffer, worker.PeekOutput() as TensorFloat, destination);

        // Execute all the commands in the command buffer
        Graphics.ExecuteCommandBuffer(commandBuffer);

        // Clear the input tensor to free its memory.
        inputTensor.Dispose();
    }

    void OnDisable()
    {
        worker.Dispose();
    }
}

For more information, refer to: