A solution for matching and detecting copied videos, developed by AImageLab and Facebook AI Research, is now being used in production scale at Facebook to detect harmful content.
Known as LAMV or TMK (Temporal Matching Kernel), this video-matching technology has been developed by research scientists at Facebook AI Research and by L. Baraldi and R. Cucchiara as part of a collaboration between AImageLab and FAIR. The algorithm is capable of producing a compact and temporal-aware signature of a video, which can be used for video copy detection and video retrieval. The solution combines match kernels, Fourier Transforms and end-to-end learning.
See the official announcement on the Facebook newsroom website, and the Github repository with the source code.
You can also read more details about this on the CVPR 2018 paper.