Face Mask Detection using Retinanet

About The Project

This project aims to create a Face Mask Detection model to visually detect facemasks on images and videos. We operate with 3 labels:

• with_mask
• without_mask
• mask_weared_incorrect

The dataset contains approximately 2500 hand-collected and hand-labelled images.

Results:

Models	mAP	with_mask	without_mask	mask_weared_incorrect	FPS (RTX 3060 Ti + CUDA)
ResNet50	68%	81%	67%	56%	~20
ResNet152	66%	81%	65%	52%	~12

How good are the models? This good:

Table of Contents

1. About The Project
2. Getting Started
   • Installation
3. Usage
• Structure
• Example
4. Contact

Getting Started

Installation

1. Clone the repo

git clone https://github.com/DvdNss/FaceMaskDetection

2. Install requirements

pip install -r requirements.txt

3. Clone PyTorch-Retinanet

git clone https://github.com/yhenon/pytorch-retinanet.git

Usage

Structure

• dataset/: contains datasets files
• retinanet/: contains retinanet scripts
• model/: contains models
• resources/: contains readme and webapp images
• annots_to_csv.py: script for datasets conversion to csv
• file_conversion.py: script for png conversion to jpg
• device.py: script for device detection (gpu or cpu)
• precompute_dataset.py: script for dataset precomputing
• app.py: streamlit webapp

Example

1. Convert datasets to csv file using annots_to_csv.py

python annots_to_csv.py --train_dataset path_to_train_dataset --valid_dataset path_to_valid_dataset --output_path path_of_outputs

2. Train a given model using pytorch-retinanet/train.py

cd pytorch-retinanet
python train.py --dataset csv --csv_train path_to_train_csv  --csv_classes path_to_class_csv  --csv_val path_to_valid_csv --depth depth_of_resnset --epochs number_of_epochs

3. Evaluate a given model using pytorch-retinanet/csv_evaluation.py

cd pytorch-retinanet
python csv_validation.py --csv_annotations_path path_to_val_annots --model_path model_path --images_path path_to_val_img --class_list_path path_to_labels

4. Visualize result using pytorch-retinanet/visualize_single_image.py

cd pytorch-retinanet
python visualize_single_image.py --image_dir image_dir_path --model_path model_path --class_list labels_path

5. Use the interface (webcam or images)

streamlit run app.py

Contact

David NAISSE - @LinkedIn - [email protected]