An important objective of video hosting and sharing platforms is to ensure that content is uploaded by the true owner or with due permission of the true owner. To ensure that unauthorized content is not uploaded, such platforms match uploaded videos against a database of original videos. A frequently encountered type of match-avoiding transform, known as framed video (or content), shrinks the original and valuable content into a small area of the video and covers the remaining portions of the video with unrelated or irrelevant patterns. This disclosure describes techniques to automatically generate large, labeled volumes of training data for framed-content types of match-avoidance strategies. The training data can be utilized to train machine learning models to more reliably detect content transforms that utilize framed content.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Pavetić, Filip; Ashirmatov, Bakhodir; and Kandpal, Mayank, "Automated Generation of Datasets of Framed Video Content", Technical Disclosure Commons, (March 18, 2021)