Computer Vision News - November 2021

2 Summary Poster Presentation 34 Best of ICCV 2021 negatives. We want things that are together in the original movie to be close to each other in that space, and far from the things that were artificially created. ” The team create metrics for the proxy task to understand whether the method is working or not. By proposing several baselines, they show that the method works to solve the proxy task. However, they need to test it on real data. To do this, they use a website called EditStock , in which aspiring editors can test their abilities on real movies. It has raw data from professionally produced independent movies, which they can download and use to create edited scenes with their algorithm. “ It is a difficult problem to measure because two different editors may come up with two different answers and both may be right, ” Alejandro tells us. “ We decided to ask people whether the edited movies made sense or not, so we designed a human study and the results showed that our method, while still far from being a professional editor, indeed works compared to all the baselines and compared to doing ranking. We have the first step towards a fully automated editor! ” Given a pair of shots from two different cameras, Alejandro and his team propose the first automated solution to this problem