Project Information

This repository contains the source code, training scripts, and contents that is used to build, deploy the Capstone Project and the Conference papers.

Capstone Project

Real-Time Covid-19 Face mask detection: Edge Intelligence with Convolutional Neural Network (CNN) and Single Shot Detectors

Paper can be found here: ./papers/(CSC3001_Capstone_Report__Final_Trimeter__Wilbur_Lua_Kai_Heng(1901839)

Youtube Link to Full Demonstration here

Conference (Accepted into 2022 IEEE 5th International Conference on Pattern Recognition and Artificial Intelligence)

Deep Learning in IoT and Edge Computing for Safety Applications

Paper can be found here: ./papers/PRAI.pdf

Motivation

• Current "Traditional" detectors (E.g. VGG-16, VGG-19, ResNet-50) have high amounts of trainable parameters leading to high computations, memory and power consumption (CPU, GPU, Network).
• Contribute to Public Health and Safety applications by researching, experimenting and deploying an FMD system that can be deployed on the Edge (NVIDIA Jetson Nano)
• Model Optimization, using Gradual Magnitude Pruning (GMP) algorithm to prune, quantize and re-train model using Sparsified transfer learning.
• Edge Intelligence, bring Deep Learning onto Internet of Things (IoT) devices.

Explaination of Source Code

Repository Breakdown

1. EfficientDet deployment codes are all under ./EfficientDet
2. YOLOv5 deployment codes are all under ./yolov5-master

Model training script are placed in /scripts/<Script_name>

1. YOLOv5s Mask Detection Script -> /scripts/YOLOv5_TrainingScript.ipynb
2. EfficientDet Mask Detection Script -> /scripts/EfficientDet_TrainingScript.ipynb
3. Model Compression and Sparse Transfer Learning Script -> /scripts/Sparse_TransferLearning.ipynb

(Required) Installing PyTorch, Torch vision and OpenCV

Scripts required to install dependencies on Edge Device - NVIDIA Jetson Nano

1. PyTorch installation Script -> /scripts/PyTorch_Torchvision.txt
2. TorchVision installation Script -> /scripts/PyTorch_Torchvision.txt
3. OpenCV installation Script (Requires to make SwapFile and Compile on-device) -> /scripts/OpenCV.txt

Miscellaneous scripts

These are the scripts I wrote to help with Data-preprocessing.

1. Utils ./utils/*

How to run the models on the Edge

YOLOv5s model

cd ./yolov5-master
python3 detect.py --weights ./weights/CMFD_Best02.pt --source 0 

EfficientDet model

cd ./EfficientDet
python3 efficientdet_test_videos.py