Microsoft’s MS Celeb dataset of face images has been quietly taken down by the company.
The open source dataset has been used to train and test facial recognition systems, with major names like IBM, Panasonic, Alibaba, and Hitachi having used it for their own purposes. Ostensibly a collection of Public Domain celebrity photos, MS Celeb also contained images of people like security journalists. Some of the featured faces, such as Surveillance Capitalism author Shoshana Zuboff and former FTC privacy commissioner Julie Brill, appear to have been included as a kind of perverse joke.
As with other major tech companies, the use of facial recognition technology has become a sensitive issue for Microsoft. The company’s president has repeatedly called for government regulation of the technology, and recently asserted that Microsoft refused to sell its biometric technology to a law enforcement agency in California on principle. But the company’s technology has also been implicated in the Chinese government’s repressive state surveillance apparatus, with a security researcher having recently said he found a reference to Microsoft’s facial recognition technology in code developed by SenseNets, one of the Chinese government’s tech suppliers. (Microsoft has denied knowledge of this use of its technology.)
As Financial Times reports, Microsoft says that the MS Celeb database was “intended for academic purposes,” and that the program has been discontinued “because the research challenge is over.” The company has also sought to distance itself from the operation of the database, asserting, “It was run by an employee that is no longer with Microsoft and has since been removed.”
Nevertheless, it was open source, meaning the database has already been downloaded many times by many users, all of whom can still use their own copies.
June 6, 2019 – by Alex Perala