Manage memory

As a Sentis user you are responsible for calling Dispose on any worker, inputs and sometimes outputs. You must call Dispose on outputs if you obtain them via worker.CopyOutput or if you take ownership of them by calling tensor.TakeOwnership.

Note: Calling Dispose is necessary to properly free up GPU resources.

For example:

public void OnDestroy()
{
    worker?.Dispose();

    // Assuming model with multiple inputs that were passed as a Dictionary
    foreach (var key in inputs.Keys)
    {
        inputs[key].Dispose();
    }

    inputs.Clear();
}

You don't need to call Dispose for the following:

  • Tensors that you receive via the worker.PeekOutput call.
  • CPUOps, GPUPixelOps, GPUComputeOps and GPUCommandBufferOps objects you create with no allocator.

Additional resources