Structured Prediction Models in Computer Vision | Efficient Convex Relaxation of Mixture Regression

Опубликовано 7 сентября 2016, 17:12

TALK 1 - Structured Prediction Models in Computer Vision Abstract: I'll present a summary of our recent work on using modern machine learning methods to solve computer vision problems. This essentially consists of using structured prediction models like max-margin structured estimators and conditional random fields. The computer vision problems we will discuss include graph matching, shape classification and object categorization. TALK 2 - Efficient Convex Relaxation of Mixture Regression with Application to Motion Segmentation Abstract: We give a semidefinite relaxation for maximum a posteriori estimation of a mixture of regression models. In addition we show how the semidefinite program can be exactly solved by a fast spectral method. We compare the proposed technique against Expectation-Maximization for synthetic problems as well as for problems of motion segmentation in computer vision, with promising results.