Simulated Masked Face Recognition Dataset (SMFRD)

SMFRD (or Simulated Masked Face Recognition Dataset) is a dataset of masked faces intended to enable facial recognition systems to identify the individuals behind the masks.

Released in March 2020 by researchers at Wuhan University in China, the set is a derivative of the Labeled Faces in the Wild (LBW) dataset, with facemasks superimposed. LBW was the first dataset to use facial images scraped from websites and applications.

Released at the height of the COVID-19 pandemic, SMFRD was seen as helpful to limiting the spread of the pandemic in China and is freely available to industry and academia.

Risks and harms 🛑

The Simulated Masked Face Recognition Dataset has raised concerns about privacy violations and its potential misuse in surveillance systems, thereby potentially limiting human rights and civil freedoms.

Transparency and accountability 🙈

The Simulated Masked Face Recognition Dataset (SMFRD) is seen to suffer from multiple transparency limitations.

• Lack of clear consent. It is unclear if and how consent was obtained from individuals whose images were used or simulated in the dataset.

• Limited information on data generation. The exact methods used to simulate masked faces may not be fully disclosed, making it difficult to assess the dataset's representativeness and potential biases.

• Inadequate documentation of intended use. The dataset's intended applications and potential misuses are not clearly outlined.

Research, advocacy 🧮

• Peg. K., Mathur A., Narayanan A. (2021). Mitigating Dataset Harms Requires Stewardship: Lessons from 1000 Papers