Hi Shery,
Could we get a bit more information on your setup? What compiler are you using? Does the armhf have Neon? ArmNN needs Neon, and while your armhf may well have it, you may need to make sure that knowledge is passed thru the compiler to your program.
Can you get any of the simple samples/examples working on your system?
Cheers,
Ben
Thanks Ben for your response.
I am using PyArmnn on docker container(20.04 Focal Fossa).
Platform OS: Debian Stretch9
Processor: arm v7l with Neon Accelarator
I wanted to run fire detection model example provided on pyarmnn repo.
I have resolved this error. It arised due to multiple versions of same libs like libarmnn,libtfliteparser etc.Once all the required libs had proper version installed( 25.0), it started accessing CpuAcc and CpuRef.
Pontential Bug in PyArnnn examples: fire detection using predict_pyarmnn.py file had some bug as it was calling an attribute that wasn't in the Pyarmnn documention.
AttributeError: module 'pyarmnn' has no attribute 'from_output_tensor'
Code that had error: output, output_tensor_info =ann.from_output_tensor(output_tensors[0][1])
change the above code line to the following attribute:
inference_results = workload_tensors_to_ndarray(output_tensors)
- Ben, please have a look at the file for this bug.If you can reproduce this at your side as well.
Gracias and waiting for your response.