The 3rd YouTube-8M Video Understanding Challenge

We took part in the Karggle competitions, the 3rd YouTube-8M Video Understanding Challenge. The award is 25,000USD and three members of us are working on it now. This project is hosted by Google Research as a part of the International Conference on Computer Vision (ICCV) 2019.

We took part in the Karggle competitions, the 3rd YouTube-8M Video Understanding Challenge. The award is 25,000USD and three members of us are working on it now. This project is hosted by Google Research as a part of the International Conference on Computer Vision (ICCV) 2019.

The mission of this project is to find and share special video moments. This technology is called temporal concept localization within video, and Google Research can use our solution to advance the state of the art in this area.

The contribution of this project:
In most web searches, video retrieval and ranking is performed by matching query terms to metadata and other video-level signals. However, we know that videos can contain an array of topics that aren’t always characterized by the uploader, and many of these miss localizations to brief but important moments within the video. Temporal localization can enable applications, such as improved video search (including search within video), video summarization and highlight extraction, action moment detection, improved video content safety, and many others.

If successful, our new machine learning models will significantly improve video understanding for all, by not only identifying the topics relevant to a video, but also pinpointing where in the video they appear.