Alexei Efros
Fri 10 Jul 2015, 15:00 - 16:00
4.31/33

If you have a question about this talk, please contact: Steph Smith (ssmith32)

Most modern visual understanding approaches rely on supervision by word labels to achieve their impressive performance.  But there are many more things in our visual world than we have words to describe them with.  Using language as supervisory signal risks missing out on much of this visual subtlety.  In this talk, I will describe some of our recent efforts to bypass the "language bottleneck" and instead use information that is already in the data, such as spatial context and visual consistency, to help in visual understanding, visual correspondence, and image retrieval.