Vidrovr is new technology that can comb through video files and identify what’s happening in a video. As a result, video content can be easily searched or automatically linked to other related material.
Reporting by Reuben Stern, Rachel Wise and Jon Doty
- Vidrovr was created by Joe Ellis and Dan Morozoff as a spinoff to a system they built at Columbia University called News Rover. According to Ellis, New Rover was a system that “recorded 100 hours of television news a day, chopped up the hourlong programs into specific topic based segments, and then extracted information like … text on screen, people on screen, what’s appearing on screen, what people are saying.” It then sorted all of the information into larger news events, which could be searched and filtered by specific topic or person.
- The Vidrovr system combines speech and facial recognition, enabling it to identify when someone well-known is speaking. Its machine-learning algorithm can also extrapolate that same information from context clues such as environment.
- Vidrovr’s system sorts detected information into several categories that include recognized persons, on-screen text and graphics, scenes detected and tags for each frame of a video.
- NYC Media Lab Combine, which is described as a “launchpad accelerator for early stage startup teams from New York City universities,” held a demo day in April 2016. Video of the Vidrovr team’s presentation can be found here.
- Vidrovr charges a fee to process each batch of video and send back the related metadata. To use Vidrovr’s other technologies like automated recommendation and publishing, clients pay upfront to purchase a license.
Rachel Wise is an editor at the Futures Lab at the Reynolds Journalism Institute and co-producer of the weekly Futures Lab video update.