Offline AI Model Conversion Toolkit (Acuity Toolkit) Overview

Introduction to Offline AI Model Conversion Toolkit

The offline AI model conversion toolkit allows user to convert their customized model into ‘.nb’ format which can be loaded directly into AmebaPro2 via example codes.

There are 2 ways to install the toolkit, installation of toolkit via docker image is highly recommended.

For installation via docker image, kindly follow the instructions provided in Acuity Toolkit Docker Installation Guide. You may also refer to this tutorial video for guidance.

For manual installation and user guide, please refer to the Toolkit Manual Installation and User Guide to install the toolkit. Kindly follow the steps provided in the guide to ensure successful model conversion.

Eligibility

Note

​To access offline AI model conversion tools, please contact AmebaAIoT@realtek.com with the subject line “Offline AI Model” using official company, institution, or educational organization email account. Please include the organization name, GitHub username, a brief description of your project, this will help us to verify your affiliation and process your inquiry more efficiently.

Once approved, please sign in to your GitHub account to download the files

Brief Comparison between Offline and Online Conversion Toolkit

User will have access to the following exclusive features available on Offline Toolkit. For every new NN model example released in the SDK, the conversion support will also be primarily given to the Offline Toolkit users first.

Features

Offline Toolkit

Online Conversion Upload

Support for YOLOv9-tiny

Yes

No

Support for MobileNetV2

Yes

No

Automation development on Docker/Linux

Yes

NA

Acuity Supported AI Framework

AI Framework

Import File Format

Caffe

.caffemodel

TensorFlow

.pb

TensorFlow Lite

.tflite

Darknet

.cfg

ONNX

.onnx

PyTorch

.pt

Keras

.h5

Note

No quantization is needed on Acuity Networks converted from ONNX, TensorFlow and Tensorflow Lite models that have been quantized. Kindly note that per-channel quantized models are NOT supported on Pro2 NPU, please ensure that your model is per-tensor quantized.

Tip

For PyTorch framework, you are highly recommended to export your file in .onnx format first to ensure successful conversion.

Reference: ACUITY Toolkit User Guide

AmebaPro2 SDK AI Model Application Types

Currently, the FreeRTOS and Arduino SDK provides several deployed models. They are listed in following table:

Category

Model

Repository

Object detection

Yolov3-tiny
Yolov4-tiny
Yolov7-tiny

https://github.com/AlexeyAB/darknet

Object detection

YOLOv7-tiny-pt

https://github.com/WongKinYiu/yolov7

Face detection

SCRFD

https://github.com/deepinsight/insightface/tree/master/detection/scrfd

Face Recognition

MobileFaceNet

https://github.com/deepinsight/insightface/tree/master/recognition

Sound classification

YAMNet

https://github.com/tensorflow/models/tree/master/research/audioset/yamnet

Reference: NN Model Zoo

AmebaPro2 SDK NN MMF examples

Please refer to the application note for the usage of NN MMF example with VIPNN module.