Medical Masks Dataset - 167, 880 Images for Face Mask Detection and Computer Vision

Overview
The Medical Masks Dataset is a large-scale, manually annotated collection of images produced by Unidata, designed to train and evaluate face mask detection models, mask detectors, and face recognition systems. The dataset consists of 167,880 images covering 41,970 unique individuals captured in 4 distinct states of medical mask usage — making it one of the largest datasets of this kind available for computer vision and deep learning research.
The dataset supports a wide range of detection tasks: from binary mask classification to fine-grained segmentation masks and object detection in real-world conditions. Images feature varied lighting, angles, and demographics to ensure robust model generalization across challenging environments.

Mask States
Each person in the dataset is photographed in all 4 mask-wearing positions:
1. No mask
2. Mask on chin
3. Mask covering mouth only
4. Mask fully on (covering mouth and nose)
This 4-state structure enables detection models to distinguish not just between masked and unmasked individuals, but to identify partial or improper mask usage — critical for safety compliance systems and surveillance in healthcare and public settings.

Dataset Composition
- Total images: 167,880
- Number of people: 41,970
- Images per person: 4 (one per mask state)
- Image format: JPG
- Data collection method: crowdsourcing platforms

Subject Demographics
The dataset covers both male and female participants spanning a wide age range (min = 18, max = 72), collected across more than 20 countries across multiple continents.
Gender distribution includes both male and female subjects. Age breakdown by group:
- Under 18: 729
- 19–25: 18,431 (largest group)
- 26–32: 11,900
- 33–39: 6,737
- 40–46: 2,713
- 47–53: 966
- 54–59: 296
- 60–66: 133
- 67+: 60

Annotations & Metadata
Each image is labeled with structured metadata, including:
- Subject ID
- Gender
- Age
- Country of origin
The annotated data enables training of both classification and segmentation models. Instance segmentation annotations and detailed bounding box labels support a variety of detection tasks, from standard object detection pipelines to more complex segmentation tasks using modern deep learning architectures.

Use Cases
- Healthcare & Public Safety. Hospitals and clinics use this dataset to train detection systems that monitor mask usage and enforce safety compliance in real time. The dataset's detailed annotations help build reliable mask detectors for institutional settings.
- Surveillance & Security. Security systems use the dataset to develop models that identify individuals wearing or not wearing surgical masks in public spaces. Manually annotated data supports detection models with the highest accuracy in crowded environments, including transportation hubs and retail locations.
- Computer Vision Research. Researchers use the dataset to experiment with deep learning methods for face recognition and mask detection tasks. The training set includes segmentation masks and supports instance segmentation workflows, making it well-suited for benchmarking proposed models against real-world conditions.

Compliance & Security
The dataset contains real-world biometric data collected from actors under controlled conditions. All data complies with GDPR and relevant data protection regulations. Storage is hosted on AWS cloud infrastructure certified to ISO 27001 and ISO 27701 standards.

Technical Specifications
Images are provided in JPG format, with 4 images per person covering all mask states. The dataset contains 167,880 images across 41,970 subjects, collected via crowdsourcing platforms. Each record is labeled with metadata: subject ID, gender, age, and country of origin. Participants range in age from 18 to 72 years, with both male and female subjects represented.

Summary
The Medical Masks Dataset is a high-quality, manually annotated image collection purpose-built for face mask detection, object detection, and face recognition research. With 167,880 images across 4 mask states, rich demographic diversity spanning 20+ countries and age groups from 18 to 72, and structured metadata per subject, it provides a comprehensive training set for building detection models that perform with the highest accuracy in real-world applications — from hospital compliance systems to large-scale surveillance and biometric security pipelines.