Read output from a model asynchronously

After you run a model, you can read the output from a model asynchronously. This avoids Sentis blocking the main thread while it waits for the model to finish running then downloads the output data to the CPU.

Follow these steps:

1. Get a reference to the tensor output data using PeekOutput.
2. Use the Tensor.PrepareCacheForAccess method and set the blocking parameter to false.
3. Use Tensor.PrepareCacheForAccess again to check if the data is complete.
4. The method returns true when the data is complete. You can then access the data.

Refer to the AsyncReadback/AsyncReadbackCompute example in the sample scripts for a working example.

Additional resources

• Tensor fundamentals
• Use output data
• Get output from any layer