Alessandro Bergamo Ph.D. Thesis Defense

Methods for Efficient Object Categorization, Detection, Scene Recognition, and Image Search, 213 Sudikoff at 11 a.m.

August 21, 2014
11 am - 1 pm
Location
213 Sudikoff
Sponsored by
Computer Science Department
Audience
Public
More information
Shannon Stearne

In the past few years there has been a tremendous growth in the usage of digital images.
Users can now access millions of photos, a fact that poses the need of having methods that can efficiently and effectively search the visual information of interest.

In this thesis, we propose methods to learn compact image representations, enabling accurate image recognition with very efficient classification models. The entries of our vectorized representations are the output of a set of basis classifiers evaluated on the photo, which capture the presence or absence of a set of high-level visual concepts.

We will propose two different techniques to automatically discover the visual concepts and learn suitable basis classifiers from a given labeled dataset of pictures. We describe several strategies to aggregate the outputs of basis classifiers evaluated on multiple subwindows of the image in order to handle cases when the photo contains multiple objects and large amounts of clutter. We extend this framework for the task of object detection, where the goal is to spatially localize an object within an image.

Since generating rich manual annotations for an image dataset is a crucial limit of the current state of the art in object localization and detection, in this thesis we also propose a method to automatically generate training data for an object detector in a weakly-supervised fashion, yielding considerable savings in human annotation.

With an exaustive set of experiments, we empirically show that our representations are able to effectively encode new unseen pictures and produce state-of-the art results in conjunt with cheap classification models on the tasks of object recognition, scene recognition, object-class search, and object localization.

 

Location
213 Sudikoff
Sponsored by
Computer Science Department
Audience
Public
More information
Shannon Stearne