Systems Science Seminar

Towards Holistic Scene Understanding: Hybrid Semantic Segmentation & Articulated Object Modeling

Roozbeh Mottaghi

Ph.D candidate
University of California, Los Angeles, Department of Computer Science
Monday, April 29, 2013
2:00pm - 3:30pm
1311 EECS

Add to Google Calendar

About the Event

Recent trends in semantic image segmentation have pushed for holistic scene understanding models that also reason about complementary tasks such as scene classification and object detection. In the first part of the talk, I will describe a hybrid human-machine scene understanding model. In this work, we are interested in understanding the roles of different cues in aiding semantic segmentation. Towards this goal, we “plug-in” human subjects for each of the various components in the model to show how much “head room” there is to improve semantic segmentation. The second part of the talk will be about the models we have developed to better capture deformations of articulated objects. The performance of current object detectors usually degrades for highly flexible objects. In this talk, I will explain how we overcome this shortcoming to achieve the state-of-the-art performance on difficult object detection benchmarks such as PASCAL VOC.

Additional Information

Contact: Ann Pace

Phone: 763-5022


Sponsor(s): University of Michigan

Open to: Public