Hardware Acceleration with Ethos-U on DEBIX Model C

2024.9.18by debix.io

Follow the official NXP documentation, i.MX_Machine_Learning_User's_Guide.pdf, to run the demo provided in 9.2.3.


1. Copy the required files and compile the model for Ethos-U using the vela tool.

cd /usr/bin/ethosu/examples/

cp ../../tensorflow-lite-2.11.1/examples/labels.txt ./

cp ../../tensorflow-lite-2.11.1/examples/grace_hopper.bmp ./

vela ../../tensorflow-lite-2.11.1/examples/mobilenet_v1_1.0_224_quant.tflite


2. Run the model through the TFLite inference engine using Ethos-U Delegate.

cd /usr/bin/tensorflow-lite-2.11.1/examples/

./label_image -m ../../ethosu/examples/output/mobilenet_v1_1.0_224_quant_vela.tflite 

--external_delegate_path=/usr/lib/libethosu_delegate.so