KaggleFaceMaskDetection

Introduction

This is a PyTorch project using Faster RCNN for 2-class face mask detection.

For Faster RCNN tutorial, please see: https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html

Dataset Description

Kaggle face mask detection dataset: https://www.kaggle.com/andrewmvd/face-mask-detection

• contains 853 images
• each image is accompanied by an annotation file, including multiple bounding boxes and labels
• 3-classes annotation is available: with_mask, without_mask, mask_weared_incorrect (not used in this project)

Folder Structure

FaceMaskDetection

|-- data
|    |-- original_data
|    |    |-- images
|    |    |-- annotations
|-- utilities
|    |-- coco_utils
|    |-- data_utils
|    |-- train_eval
|-- output
|-- model
README.md
requirements.txt
setup.py
train.py
test.py

Environment Setup

The project was written under the following environment:

• Python==3.7
• torch==1.4.0
• torchvision==0.5.0
• pycocotools==2.0.2
• nvidia-ml-py3
• xml
• PIL

Before running any other code, please run the setup.py in the root folder first, so to setup the file paths properly.

Data Preprocessing

1. Download the dataset and put the images and annotation files under the corresponding folders
2. go to utilities--> data_utils--> split_dataset.py
3. now check the data folder, there should be two new sub-folders now: train and test

Train and Test

• For training, please run train.py so to generate the model; or you can simply download the pre-trained model from: https://drive.google.com/drive/folders/1UuU5up0DQfRuafdbZrayPdRwpVvJOYYK?usp=sharing
• the testing results will be written into the output folder, here's an example of prediction: