alexlorenzo / masked-faces-detection

Detect masked faces using CNNs

Mask Detector using YOLOv3

Ultralytics	On mask detection

The purpose of this repository its to train your own YOLO model on new datasets for that we provided mask/no mask datasets.

Requirements

To manage package and dependency easily I used the poetry package. Simply use the following command line once you have installed poetry, it will create automatically a .venv folder.

$ poetry install 

If you have issues installing poetry, check that:

$ poetry config --list

returns:

cache-dir = "/Users/user/Library/Caches/pypoetry"
virtualenvs.create = true
virtualenvs.in-project = true
virtualenvs.path = ".venv"

If you want to execute the code right away, I added a notebook *YOLOv3 Startkit", which you can run in Kaggle.

Repository structure

masked-faces-detection
└── yolov3: package from ultralytics yolov3 
    ├── weights (include weights from ultralytics or Kaggle datasets)
└── mask: include datasets (ex: in Kaggle datasets)
    ├── images
    |   └── include jpg files
    ├── labels
        |   └── include txt files
    ├── train.txt
    ├── test.txt
    ├── valid.txt
└── notebooks: some useful notebooks
└── Mask_Detection_Yolov3_ultralytics.ipynb: implementation using Mask Datasets
└── pyproject.toml
└── poetry.lock
└── .gitignore
└── README.md

Reproduce our result

I recommend to follow the Customer Train Custom Data from Ultralytics. If not, follow the instruction in the notebook from this repository.

Datasets

All datasets have been found by scrapping Google images and from the real/fake datasets from Kaggle. We have labelled all images using CVAT package and exporting our result in the YOLO format. You can find the pre-trained weights (Pytorch format) as well as the data on the Kaggle.