qu-ngx / nufit-ml

Machine Learning Backend for NuFit App - GSC24.

NuFit - Machine Learning Backend by Quang Nguyen (Completed)

Overview:

• This repo is a small demo Computer Vision app deployed an AI/ML model to detect different kind of food. This repo is a subsystem/backend of NuFit
• The implementations took an average of 180 hours of testing different approaches to complete the first full deployment on Flutter.
• Machine Learning Model: SSD-MobileNet-V2-300x300 (Also work with SSD-MobileNet-V2-FPN-Lite-320x320)
• Method of training: Transfer Learning on trained model with COCO2017 set.
• Input/Output layer: Kaggle

Folder Structures:

• There are 3 main folders here with backend and object_detection_ssd_mobilenet being the main components.
• backend included Jupyter notebook to help training and setting up an online virtual env without having to install locally on personal laptop and computers, labelimg for image labelling, and model quantizer for tflite file size reduction (32 bit down to 8 bit)
• object_detection_ssd_mobilenet - The main app running AI/ML or Food Detection (See the image Demo below)
• There are indepth uses and documentations inside the codes for easy run.
• For guide to install labelimg: labelimg

How does it work?

• Since this is a completed seperate part, this part alone can be run independently as an app.
• There are 8 types of food that this AI/ML model can detect: french_fries

 sausage
 grilled_chicken
 fish
 scrambled_egg
 pasta
 lettuce
 cantaloupe

• Trained ML model can detect 8 types of food above

Demo & Run App Demo:

• Installation guide & Run Guide: [Coming Soon]

Machine Learning App Demo:

• For custom deployment or other model deployment: Please refer to flutter-tflite