Torralba A, 2011, "Predicting the future" Perception 40 ECVP Abstract Supplement, page 2
Predicting the future
In this talk I make a link with computer vision and recent techniques for addressing the problem of predicting the future. Some of the representations to address this problem in computer vision are reminiscent of current views on scene understanding in humans. When given a single static picture, humans can not only interpret the instantaneous content captured by the image, but also they are able to infer the chain of dynamic events that are likely to happen in the near future. Similarly, when a human observes a short video, it is easy to decide if the event taking place in the video is normal or unexpected—even if the video depicts an unfamiliar place for the viewer. This is in contrast with work in computer vision, where current systems rely on thousands of hours of video recorded at a single place in order to identify what constitutes an unusual event. I discuss techniques for predicting the future based on a large collection of stored memories. I show how—relying on large collections of videos using global images features such as the ones used to model fast scene recognition—it is possible to index events stored in memory similar to the query, and how a simple model of the distribution of expected motions can be built. Consequently, the model can make predictions of what is likely to happen in the future as well as evaluate how unusual is a particular event.
These web-based abstracts are provided for ease of seaching and access, but certain aspects (such as as mathematics) may not appear in their optimum form. For the final published version of this abstract, please see
ECVP 2011 Abstract Supplement (complete) size: 2206 Kb
[Publisher's note: The abstracts in this year's ECVP supplement have been published with virtually no copy editing by Pion, thus the standards of grammar and style may not match those of regular Perception articles.]