When: Wednesday, March 28, 2018
Where: PGH 563
Time: 11:00 AM

Making Consensus-based Decisions in Person Re-identification using a Graph-based Approach

Speaker: Arko Barman

Person re-identification is a challenge in video-based surveillance where the goal is to identify the same person in different camera views. In recent years, many algorithms have been proposed that approach this problem by designing suitable feature representations for images of persons or by training appropriate distance metrics that learn to distinguish between images of different persons. Aggregating the results from multiple algorithms for person re-identification is a relatively less-explored area of research. In this work, we formulate an algorithm that maps the ranking process in a person re-identification algorithm to a problem in graph theory. We then extend this formulation to allow for the use of results from multiple algorithms to make a consensus-based decision for the person re-identification problem. The algorithm is unsupervised and takes into account only the matching scores generated by multiple algorithms for creating a consensus of results. Further, we show how the graph theoretic problem can be solved by a two-step process. First, we obtain a rough estimate of the solution using a greedy algorithm. Then, we extend the construction of the proposed graph so that the problem can be efficiently solved by means of Ant Colony Optimization, a heuristic path-searching algorithm for complex graphs. While we present the algorithm in the context of person re-identification, it can potentially be applied to the general problem of ranking items based on a consensus of multiple sets of scores or metric values.

Bio:

Arko Barman received his Bachelor’s degree in Electrical Engineering from Jadavpur University, India in 2009 and his Master’s degree in Signal Processing from Indian Institute of Science in 2011. Subsequently, he worked at Broadcom Corporation for over a year, developing software for 4G LTE chips, and served as an Assistant Professor at a 4-year engineering college in India. Arko is presently a PhD student and Teaching Assistant at Department of Computer Science at University of Houston. His research interests include Computer Vision, Machine Learning, Data Mining and Heuristic Optimization Algorithms. He has taught or has assisted in teaching a diverse range of courses.

Enhanced Multiple Sclerosis Lesion Detection through MRI-based Augmented-Reality

Speaker: Dan Biediger

Multiple sclerosis (MS) is an autoimmune disease of the central nervous system that causes damage to the insulating myelin sheaths around the axons in the brain. It affects over 2.5 million people worldwide, with 200 new cases per week in the United States. The disease progresses at different rates in different people and can have periods of remission and relapse. A fast and accurate method for evaluating the number and size of MS lesions in the brain is a key component in evaluating the progress of the disease and efficacy of treatments. The normal method for imaging MS lesions is utilizing Magnetic Resonance Imaging (MRI). MS lesions are more apparent in certain MRI modalities and the segmentation of these lesion usually requires careful interpretations by trained physicians. Due to the structure of the brain and the appearance of MS in MRI images, the manual segmentation process is slow, difficult, and the results can be somewhat subjective. Physicians must recognize local contrasts in the brain while keeping track of the location in the brain and the corresponding probable tissue type. To decrease the cognitive load on the physicians and improve this process, we propose an augmented reality display of the MRI data using the Microsoft HoloLens. Because MRI data are inherently three-dimensional, it is natural to present and interact with these data in three-dimension. This provides location and orientation clues not available in flat screen formats, and allows for a more natural interaction with the data.

Bio:

Dan Biediger is a second year PhD student with a background in engineering as well as computer science. He is interested in applying mixed reality to problems of visualization, education, simulation, and training.

Deep Imbalanced Attribute Classification using Visual Attention

Speaker: Nikolaos Sarafianos

Visual attributes whether facial or human-body are imbalanced in nature. Bald people wearing sunglasses and carrying a backpack can be more than 40 times less likely to appear in a dataset compared to more common attributes. Handling class imbalance in deep learning is a challenging problem with traditional solutions such as over-sampling, undersampling, and cost-sensitive learning failing to work well in large image datasets. In this talk, we will present a method to handle class imbalance by leveraging visual attention. Given an input image, attribute-specific regions are identified in a weakly-supervised manner to guide the network to put more attention to these regions. In addition, during training, we focus not only on the hard samples but also in those that demonstrate high variance in their visual attention masks. Extensive evaluations and ablation studies are performed on the CelebA and WIDER datasets that comprise facial and human attributes.

Bio:

Nikolaos Sarafianos received the Diploma in Electrical and Computer Engineering from the National Technical University of Athens, Greece in 2013 and is currently pursuing the Ph.D. degree in the Department of Computer Science at the University of Houston. His industry experience includes working as a research scientist intern at Amazon Alexa in 2017 and Facebook's Oculus Research group in summer 2018. His research interests include Deep Learning and Computer Vision with applications to Biometrics.