Unsupervised and Weakly-Supervised Object Discovery, Segmentation, Classification, and Clustering

Zhuowen Tu
University of California, Los Angeles
Location: 
Wachman 1015D
Date: 
Thursday, February 28, 2013 - 01:30

In the first part of the talk, I will present two recent papers (CVPR 2012) along the line of weakly supervised learning for unsupervised object discovery and an integrated framework for cancer image segmentation, classification, and clustering. We tackle the problem of common object (multiple classes) discovery from a set of images and propose an algorithm for simultaneously localizing objects and discovering object classes via bottom-up (saliency-guided) multiple class learning (bMCL).

 In the second part of the talk, I will briefly show a range of work recently developed in my group including: medical document analysis in the i2b2 challenge, object discovery using both low-rank and discriminative models, human gesture understanding, and weakly supervised learning for microscopic image segmentation/clustering.