Source-linked AI summary
MaskedFace-Net -- A Dataset of Correctly/Incorrectly Masked Face Images in the Context of COVID-19
Adnane Cabani, Karim Hammoudi, Halim Benhabiles, Mahmoud Melkemi
TL;DR
The paper addresses the lack of a large dataset distinguishing correctly from incorrectly worn masks, a gap relevant to masked-face recognition. It constructs MaskedFace-Net from FFHQ using deformable mask-to-face mapping and releases 137,016 images for mask-presence and mask-wearing analysis, while noting age coverage and use-related boundaries.
Problem
Existing masked-face datasets do not provide large-scale classification of whether detected masks are correctly or incorrectly worn.
Method
The authors map a reference mask onto FFHQ faces using facial detection, landmarks, four wearing configurations, and homography transformation.
Results
137,016 masked face images remain after filtering failed detections and erroneous landmark mappings.
Takeaways & Limitations
MaskedFace-Net supports detecting mask presence, classifying correct or incorrect wearing, and generating images with other mask types.
Takeaways & Limitations
The dataset applies raw mapping without age filtering and includes faces of all ages, while mask-wearing age limits vary by country.
Abstract
from arXiv · showhide
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.
1 Introduction and motivation
Existing masked-face datasets support mask detection, but the paper identifies a gap in large-scale data distinguishing correctly from incorrectly worn masks. It proposes CMFD, IMFD, and their combined MaskedFace-Net dataset for these analyses.
- Large masked-face datasets exist, including MAFA, RMFD, MFDD, RMFRD, and SMFRD.
- 137,016 masked face images are divided into correctly worn and incorrectly worn mask categories.The incorrectly worn category addresses bad practices and vulnerability among people such as children and older adults.
- The paper proposes CMFD, IMFD, and their combination, MaskedFace-Net, for masked-face detection.Figure 2a presents the dataset structure.
- The datasets target both detecting whether faces are masked and classifying whether detected masks are correctly worn.The stated applications include airport portals and crowds.
2 Applied mask-to-face deformable model
MaskedFace-Net is generated by deformably mapping a reference mask onto FFHQ faces using detected facial landmarks. Four mapping cases represent correct and incorrect mask wearing, followed by filtering of failed detections and mappings.
- Applied mask-to-face deformable model: FFHQ supplies 70,000 high-quality 1024 × 1024 face images with varied age, ethnicity, viewpoint, lighting, and backgrounds.The dataset is used as the base for creating correctly and incorrectly masked images.
- Applied mask-to-face deformable model: Face rectangles and 68 facial landmarks are automatically detected before mapping a conventional blue protection mask onto each face.Haar cascade classifiers detect the face region, while a shape predictor detects the landmarks.
- Applied mask-to-face deformable model: Four mappings encode correct coverage of the nose, mouth, and chin, plus three incorrectly worn configurations.Each configuration retains 12 facial key points and matches them to 12 mask key points.
- Applied mask-to-face deformable model: Homography transformation maps mask pixels onto the targeted facial areas using landmark correspondences.Figure 4 shows landmark detection and the resulting mask-to-face mappings.
- Applied mask-to-face deformable model: 137,016 images remain after removing 177 undetected faces and manually filtering erroneous landmark mappings.The final dataset contains 67,193 correctly masked and 69,823 incorrectly masked images.
- Applied mask-to-face deformable model: The raw mapping applies no filtering by parameters such as age, while FFHQ filenames preserve correspondence for related filtering.Because FFHQ contains faces of all ages, MaskedFace-Net also includes all ages.
3 Conclusion
The paper releases MaskedFace-Net as a large dataset for analyzing mask presence and wearing correctness, and describes a generation method adaptable to other mask types.
- 137,016 quality masked face images are produced and made available online.
- MaskedFace-Net supports models detecting mask presence and distinguishing correct from incorrect mask wearing.The paper identifies applications including monitoring legal compliance and generating crowd statistics.
- The described generation method permits creating masked face images with other mask types.
4 Disclaimer
The disclaimer states that the work’s contributors accept no responsibility for incidents arising from use of the MaskedFace-Net dataset or masks.
- The contributors disclaim responsibility for any incident involving use of the MaskedFace-Net dataset or masks.