Russia’s censorship agency Roskomnadzor has launched Oculus, a system for search and analysis of banned content, Russian newspaper Vedomosti reports, citing a representative of Russia’s Radio Frequency Centre that had commissioned the project.
According to the representative, the system has been launched and is performing tasks at full scale.
“The system recognises images and symbols, unlawful scenes and activities, analyses text in photo and video materials. Oculus automatically detects such wrongdoings as [materials] with extremist themes, calls for mass illegal rallies, suicide, pro-drug content, LGBT propaganda, and other,” the newspaper’s source explains.
The source adds that the system was tested in December 2022. In January 2023, it was integrated together with other instruments of Roskomnadzor’s monitoring.
According to the developers’ plans, Oculus will analyse “over 200,000 images per day (about three seconds per image)”. In the past, providers had to manually analyse 106 images and 101 videos per day. In the future, the system will learn to recognise banned material “in several shots of video fragments, in hard-to-understand handwritten texts, and in drawn content”.
In August 2022, Russian newspaper Kommersant learnt that 57.7 million rubles (€750,000) had been allocated by the Radio Frequency Centre to develop Oculus.
The centre’s Digital Technologies Director Konstantin Bulanov said in the past that at least 48 servers fitted with graphics accelerators would be needed for the system to operate in a stable manner. According to him, Oculus will search for content through neural networks that operate based on deep machine learning.
Media outlet IStories previously reported that Oculus would find photos and videos of protests and identify its participants. Furthermore, the system will monitor sharing of photoshopped pictures of Vladimir Putin.
The list of violations indicated in Oculus’ technical specifications also includes “depicting the LGBT lifestyle as attractive” and “images of individuals non conforming to the traditional of man and woman (for example, masculine female faces, men with makeup on).
