I'm working on the tensorflow lite object detection example. The app runs in CPU at startup. I want do run tflite using GPU from start. However whenever I start the app with the GPU option set, the app crashes with the above mentioned error.
The error message however seems to be misleading: If I build in a delay of 10 s before running the inference starts, everything is working fine. This is a hint that the GPU-based classifier is set up asynchronously in another thread.
Is there a way to wait and check until the TfLiteGpuDelegate is up and running before starting the classification?
I'll try to come up with a minimal working example, but the code is quite involved.