Baidu Releases Open-Source Face Mask Detection Model

China’s leading search engine and AI company Baidu released the industry’s first face detection and classification model to detect whether individuals in crowded areas are wearing masks or not. As an open-source AI tool, it will allow developers and manufacturers to develop detection software that is urgently needed as China tries to contain the novel coronavirus that has infected more than 60,000 in China.

After an extended Lunar New Year holiday, over 70% of Chinese businesses have resumed work this week. Face mask detection applications are in high demand from transportation centers, densely populated areas, residential districts, large-scale manufacturers and other enterprises.

The open-source model is based on the research and development of PyramidBox, a context-assisted single shot face detector model published in 2018. It can quickly identify faces wearing or not wearing face masks in crowds.

Based on this model, developers can develop face mask detection softwares in different scenarios with a small amount of their own data. Only basic Python programming skills are required to complete the application development. It can also be easily adjusted for mobile devices.

The model has an accuracy of 96.5%, which meets the requirement for routine inspection, according to a Baidu engineer. Developers can optimize the quadratic model using their own data to further improve the accuracy.