I tested the Image classification example given in UM11226 section 7.2 On iWG27M iMX8QM SOM development kit, with Yocto ML manifest having 4.9.51 beta kernel with Opencl support.
It works fine and predicts the results in around 300 ms but it does not uses GPU. gputop application do not show any process running.
So My questions are
1) Why the application is not using GPU for its execution? (In spite of OpenCL support).
2) What needs to be done to use the GPU support.
3) I need TensorFlow-lite object detection example which utilize GPU for its execution, where Can I find that?