M58 - A "Shape Aware" Model for Semi-Supervised Learning of Object and its Context
Abhinav Gupta, Jianbo Shi and Larry S. Davis

Learn "Syntactic + Semantic" Model from Weakly Labeled Data

Co-occurrences not sufficient

Intervening Contours

Using "Sketches" for Learning
Cost Function: Co-occurrence + Intervening Contours + Shape Similarity Iterative Formulation
Co-occurrence Co-location Shape Matching

Localization - Training

Localization - Test

Advantages of Iterations