Computer vision researchers have made a lot of progress by thinking in terms of pixels and patches, image processing, and pattern recognition. Now, we need vision algorithms that go beyond the pixels of the image to reason about the underlying scene. I'll describe work in the dissertation and beyond that shows how we can recover a rough, qualitative sense of the 3D scene from a single image, enabling computers to create pleasing 3D reconstructions from one photograph, to more accurately recognize objects, and to provide more intuitive image editing interfaces. I'll also discuss key research directions towards more coherent and robust visual systems.
Derek Hoiem received his Ph.D. from the Robotics Institute at Carnegie Mellon in 2007. His dissertation work with Alexei Efros and Martial Hebert was featured in the Economist and Slashdot and was honored with the SCS Distinguished Dissertation Award and an honorable mention for the ACM Doctoral Dissertation Award. Derek is now an assistant professor at University of Illinois at Urbana-Champaign (UIUC). His research interests include object recognition and scene-based reasoning from images.