VX TFLite

TFlite implementations from https://github.com/sravansenthiln1/armnn_tflite/ adapted to run on Verisilicon's Vivante NPU hardware platform.

Compatible with VIM3/3L, on the 4.9 and 5.15 kernel.

Install pip

sudo apt-get install python3-pip

Install necessary python packages

pip3 install numpy pillow opencv-python librosa

Install the TFLite runtime interpreter

pip3 install --extra-index-url https://google-coral.github.io/py-repo/ tflite_runtime

Create Library symlink

sudo ln /usr/lib/libOpenVX.so /usr/lib/libOpenVX.so.1

Copy the necessary library files

If you are running on VIM3:

sudo cp libs/VIM3/libtim-vx.so /usr/lib/aarch64-linux-gnu/

If you are running on VIM3L:

sudo cp libs/VIM3L/libtim-vx.so /usr/lib/

Examples

Simple Neural Networks

model	description
Sine Model	Basic Neural network TFLite model
Digit recognize Model	Digit recognization model

Audio Processing

model	description
Audio classifier Model	Audio classifier model

Image processing

model	description
Mobilenet v1 Model	Mobilenet v1 image classification model
YOLOv8n Model	YOLOv8n image detection model
Auto crop Model	Automatic document crop model