MaskedFace-Net -- A Dataset of Correctly/Incorrectly Masked Face Images in the Context of COVID-19 (2008.08016v1)

Abstract: The wearing of the face masks appears as a solution for limiting the spread of COVID-19. In this context, efficient recognition systems are expected for checking that people faces are masked in regulated areas. To perform this task, a large dataset of masked faces is necessary for training deep learning models towards detecting people wearing masks and those not wearing masks. Some large datasets of masked faces are available in the literature. However, at the moment, there are no available large dataset of masked face images that permits to check if detected masked faces are correctly worn or not. Indeed, many people are not correctly wearing their masks due to bad practices, bad behaviors or vulnerability of individuals (e.g., children, old people). For these reasons, several mask wearing campaigns intend to sensitize people about this problem and good practices. In this sense, this work proposes three types of masked face detection dataset; namely, the Correctly Masked Face Dataset (CMFD), the Incorrectly Masked Face Dataset (IMFD) and their combination for the global masked face detection (MaskedFace-Net). Realistic masked face datasets are proposed with a twofold objective: i) to detect people having their faces masked or not masked, ii) to detect faces having their masks correctly worn or incorrectly worn (e.g.; at airport portals or in crowds). To the best of our knowledge, no large dataset of masked faces provides such a granularity of classification towards permitting mask wearing analysis. Moreover, this work globally presents the applied mask-to-face deformable model for permitting the generation of other masked face images, notably with specific masks. Our datasets of masked face images (137,016 images) are available at https://github.com/cabani/MaskedFace-Net.

• The paper presents a dataset categorizing 137,016 face images into correctly and incorrectly masked classes to support mask compliance evaluation.
• It employs a mask-to-face deformable model using facial landmark detection on the FFHQ dataset to simulate realistic masked images.
• The dataset enhances AI model training for public safety by providing detailed classifications for proper and improper mask usage.

Overview of "MaskedFace-Net - A Dataset of Correctly/Incorrectly Masked Face Images in the Context of COVID-19"

The paper outlines the development and release of MaskedFace-Net, a comprehensive dataset comprising 137,016 images of masked faces. This dataset is crucial for advancing machine learning models that assess mask-wearing accuracy, particularly amidst the global context of COVID-19. The dataset differentiates between correctly masked faces and various types of incorrectly masked faces, filling an existing gap in publicly available resources for mask compliance evaluation.

Contributions

The primary contribution of this work is the establishment of MaskedFace-Net, which uniquely categorizes masked face images into correctly and incorrectly worn classes. It introduces:

• Correctly Masked Face Dataset (CMFD): Containing over 67,000 images of properly masked individuals.
• Incorrectly Masked Face Dataset (IMFD): Comprising over 69,000 images, subdivided into cases such as masks worn below the nose, covering only the mouth and chin, and worn below the chin.

Through this structural classification, the dataset enhances the granularity at which facial recognition systems can operate, a feature previously absent from existing databases.

Methodology

The dataset was generated using the Flickr-Faces-HQ (FFHQ) dataset as a base, transformed using a mask-to-face deformable model that leverages facial landmark detection. This model adapts a generic mask image to each face by matching specific landmarks to pre-determined mask positions. The paper provides detailed pseudo-code describing this transformation process, enabling reproducibility and potential customization with different mask designs.

Implications

MaskedFace-Net can significantly impact several domains:

• Public Safety and Health Compliance: Facilitating the development of systems that ensure adherence to mask mandates through accurate real-time detection of mask-wearing behaviors.
• AI Model Training: Providing a substantial and varied dataset for training advanced models capable of distinguishing between correct and incorrect mask usage.

On a theoretical level, the dataset contributes to the growing field of machine vision, particularly in improving the robustness of recognition models that understand context-specific nuances, such as the variations in mask placement.

Future Developments

The dataset sets the stage for several future research directions, including:

• Integration with real-world surveillance systems to automate compliance monitoring.
• Expansion of the dataset to include various mask types, contexts, and diverse demographic groups.
• Further refinement of the mask-to-face deformable model to enhance accuracy and applicability across different facial recognition tasks.

Overall, MaskedFace-Net stands as an essential tool for advancing research in mask compliance and facial recognition, offering valuable insights and foundational resources for developing responsive, intelligent systems in public safety contexts.