Real-World Masked Face Dataset
Amid the recent global COVID‑19 outbreak, regions severely affected (e.g., Wuhan) saw near‑universal mask usage, creating a massive pool of samples. We collected these to build the world’s largest masked‑face dataset, released publicly to support current and future public‑safety scenarios. Using this data, we design mask‑occluded face detection and recognition algorithms for closed‑community access control, upgraded facial‑recognition turnstiles at stations and airports, and attendance systems that operate under mask‑cover conditions.
Dataset description and usage context
Dataset Overview
Dataset Name
- Masked‑Face Dataset (Real‑World Masked Face Dataset, RMFD)
Initiating Institution
- Wuhan University National Multimedia Software Engineering Technology Research Center
Contact
- Contact Person: Xiong Zhangyang
- Email: x_zhangyang@whu.edu.cn
Dataset Content
-
Real‑World Masked Face Recognition Dataset
- Source: Web crawling
- Size: 525 individuals, 5 k masked faces, 90 k normal faces
- Download:
- Baidu Cloud (Password: j3aq)
- Google Drive
-
Simulated Masked Face Recognition Dataset
- Source: Existing face datasets with synthetic masks applied
- Size: 10 k individuals, 500 k faces
- Sub‑datasets:
- WebFace Simulated Masked Set
- Download:
- Baidu Cloud (Password: 77m8)
- Google Drive
- Download:
- LFW Simulated Masked Set
- Download:
- Baidu Cloud (Password: o126)
- Google Drive
- Download:
- WebFace Simulated Masked Set
Intended Use
- Design and train algorithms for mask‑occluded face detection and recognition.
- Apply to community lockdown access control, upgraded facial‑recognition turnstiles at stations/airports, and facial‑recognition attendance devices.
Performance
- Using the dataset, a multi‑granularity mask‑face recognition model achieved 95 % identification accuracy.
Download Packages
- RMFD_part_1: Directly usable.
- RMFD_part_2: Requires downloading all four archives and extracting.
- RMFD_part_3: Requires downloading all three archives and extracting.
- Links:
- Baidu Cloud (Code: xhze)
- Google Drive
Pair the dataset with AI analysis and content workflows.
Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.