BestTechVideos: Category DSP Videos

What Do Those Images Have In Common?

scoundrel — Fri, 25 Apr 2008 01:39:50 GMT

Google Tech Talks
March, 25 2008

ABSTRACT

This talk is about discovering and modeling previously unspecified, recurring themes in a given set of arbitrary images. Given a set of images, each containing frequent occurrences of objects from multiple categories, the goal is to learn a compact model of the categories as well as their relationships, for the purposes of later recognizing/segmenting any occurrences in new images. Categories are not defined by the user. Also, whether and where instances of any categories appear in a specific image is not known. This problem is challenging, since it involves the following unanswered questions. What is an object category? What image properties should be used and how to combine them to discover category occurrences? What is an efficient multicategory representation?

We will examine a methodology, developed during my postdoctoral work at UIUC. Each image is represented by a segmentation tree whose nodes correspond to image regions, segmented at all natural scales present, and edges between tree nodes capture the region embedding. The presence of any categories in the image set is then reflected in the frequent occurrence of similar subtrees within the segmentation trees. Our methodology is designed to: (1) match image trees to find similar subtrees; (2) discover categories by clustering similar subtrees, and use the properties of each cluster to learn the model of the associated category; and (3) learn the grammar of the discovered categories that compactly captures their recursive definitions in terms of other simpler (sub)categories and their relationships (e.g., containment, co- occurrence, and sharing of simple categories by more complex ones). When a new image is encountered, its segmentation tree is matched against the learned grammar to simultaneously recognize and segment all occurrences of the learned categories. This matching also provides a semantic explanation of object recognition in terms of the identified parts along with their spatial relationships.

The aforementioned methodology can also be used for identifying recurring image themes of more general kind. An example is that of extracting the stochastically repeating, elementary parts of image texture (e.g., waterlilies on the water surface, people in a crowd).

This talk will be taped by the engEDU Tech Talks Team.

Speaker: Sinisa Todorovic
Sinisa Todorovic received the joint B.S./M.S. degree with honors in electrical engineering from the University of Belgrade, Serbia, in 1994. From 1994 until 2001, he worked in the communications industry. He received the M.S. and Ph.D. degrees in electrical and computer engineering at the University of Florida, Gainesville, in 2002, and 2005, respectively. Since 2005, he holds the position of Postdoctoral Research Associate in the Beckman Institute at the University of Illinois Urbana-Champaign, where he collaborates with Prof. Narendra Ahuja. Sinisa's main research interests concern computer vision and machine learning, with current focus on unsupervised extraction and representation of visual themes recurring in images. He is the recipient of Jack Neubauer Best Paper Award 2004 for a publication in IEEE Trans. Vehicular Technology, and Outstanding Reviewer Award at the Int. Conf. on Computer Vision (ICCV) 2007. He serves as Associate Editor of Advances in Multimedia.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Companies, Science, Techtalks, Google, Lectures, DSP, Computer Science

Movie/Script: Alignment and Parsing of Video and Text Transcription

scoundrel — Mon, 07 Apr 2008 22:51:25 GMT

Google Tech Talks
March, 26 2008

ABSTRACT

Timothee Cour - Research Scientist

Movies and TV are a rich source of highly diverse and complex video of people, objects, actions and locales "in the wild". Harvesting automatically labeled sequences of actions from video would enable creation of large-scale and highly-varied datasets. To enable such collection, we focus on the task of recovering scene structure in movies and TV series for object/person tracking and action retrieval. We present a weakly supervised algorithm that uses the screenplay and closed captions to parse a movie into a hierarchy of shots and scenes. Scene boundaries in the movie are aligned with screenplay scene labels and shots are reordered into a sequence of long continuous tracks or threads which allow for more accurate tracking of people and actions across shot boundaries. Scene segmentation, alignment, and shot threading are formulated as inference in a unified generative model and a novel hierarchical dynamic programming algorithm that can handle alignment and jump-limited reorderings in linear time is introduced. We present quantitative and qualitative results on movie alignment and parsing, and use the recovered structure for tracking and naming of characters as well as retrieval of common actions in several episodes of popular TV series.

If time permits we will also present our recent results on approximate inference with eigenvalue optimization.

Speaker: Timothee Cour - Research Scientist
Timothee Cour is a fifth year PhD student at the University of Pennsylvania, Philadelphia, in Computer Science. He completed his undergraduate education at the Ecole Polytechnique in France, majoring in Computer Science and Applied Mathematics. His research advisor is Prof. Ben Taskar and he also worked closely with Prof. Jianbo Shi.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Science, Techtalks, Google, Lectures, DSP, Computer Science

A New Marriage of Brain and Computer

scoundrel — Fri, 12 Oct 2007 19:40:29 GMT

Google Tech Talks
September, 21 2007

ABSTRACT

Brain and computer were wed mid-twentieth century by the McCulloch-Pitts model neuron and Hodgkin-Huxley equations for digital firing in biological neurons. Since then, brain neurons, synapses, firings and networks have been considered analogous to electronic switches, states and circuits in classical computers. But despite extraordinary advances and bold predictions, consciousness seems ever
more elusive. On this, and other divisive issues like EEG gamma synchrony, deviations from Hodgkin-Huxley, gap junctions, dendritic
webs/hyper-neurons, anesthesia, quantum computers and clear demonstration of functional quantum coherence in warm protein assemblies, brain and computer have drifted apart. Increasing evidence
suggests that the brain-computer marriage could be spiced up, and consciousness accounted for, by exotic yet testable new approaches such as quantum computation in microtubules (Penrose-Hameroff Orch OR theory) in laterally connected input layers (gap junction dendritic web/hyper-neurons) of neuronal networks.

Speaker: Stuart Hameroff
Stuart Hameroff M.D. is an anesthesiologist, Professor of Anesthesiology and Psychology and Director, Center for Consciousness Studies at the University of Arizona in Tucson.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, DSP, Computer Science

Video Mosaicing for Non-Chronological Time Editing

scoundrel — Fri, 13 Jul 2007 17:04:18 GMT

Google Tech Talks
June 27, 2007

ABSTRACT

Non-Chronological Video Mosaicing involves the generation of a new video from an input video such that the chronological order of events is not preserved. The freedom from the chronological constraint enables multiple effects that are otherwise impossible:

The creation of dynamic panoramic video of a dynamic panoramic scene from a regular camera scanning this scene.
The generation of virtual immersive walk-through from a regular camera scanning a city street.
The summarization of video into a shorter video. This summarization is especially effective for continuously transmitting webcams.

The computation of camera ego-motion is needed for cases (1) and (2). Robust methods will be described for computing this motion in the common case of a camera moving along a straight line, e.g. in a driving car. Using these methods, city-block models can be populated by people uploading videos of their neighborhoods.

We will also present dynamic video synopsis, where most of the activity in the video is condensed by simultaneously showing several actions, even when they originally occurred at different times. These methods are especially suitable for summarization of video transmitted continuously from webcams. Embedding webcam summaries in city-block images can complete the dynamic information, missing from existing geographical descriptions. Webcams can be supplied by the community as well, already populating millions of cameras on the web.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, DSP, Computer Science

Supersymmetry, Extra Dimensions and the Origin of Mass

scoundrel — Wed, 11 Jul 2007 16:42:59 GMT

Google Tech Talks
June 18, 2007

ABSTRACT

"Supersymmetry, Extra Dimensions and the Origin of Mass: Exploring the Nature of the Universe Using PetaScale Data Analysis"

The Large Hadron Collider (LHC), scheduled to begin operation in Summer 2008, will collide protons at energies not accessible since the time of the early Universe. The study of the reactions produced at the LHC has the potential to revolutionize our understanding of the most fundamental forces in nature. The ATLAS experiment, currently being installed at the LHC, is designed to detect collisions at the LHC, to collect the relevant data and to provide a unified framework for the reconstruction and analysis of these data. This talk will review the goals of the ATLAS program and will describe the software and computing challenges associated analyzing these data. Among the relevant issues are the need to develop and maintain a unified analysis framework for use by more than 1000 scientists and the need for distributed access to large (petabyte) scale data samples, including a significant metadata component.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Math, DSP, Computer Science

Learning to Analyze Sequences

scoundrel — Sat, 05 May 2007 10:35:19 GMT

Google Tech Talks
April 12, 2007

ABSTRACT

Sequential data --- speech, text, genomic sequences --- floods our storage servers. Much useful information in these data is carried by implicit structure: phonemes and prosody in speech, syntactic structure in text, genes and regulatory elements in genomic sequences. Over the last six years, several of us have been investigating structured linear models, a unified discriminative learning approach to sequence analysis problems. I will review the approach and illustrate it with applications to information extraction and gene finding. I will conclude with a summary of other applications and current research questions.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, DSP, Computer Science

Learning Image Patch Representation for Scene Recognition

scoundrel — Fri, 06 Apr 2007 12:36:53 GMT

Google TechTalks
May 9, 2006

Le Lu

Learning Image Patch Representation for Scene Recognition, Object Tracking, and Dynamic Foreground/Background Extraction

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Math, DSP, Computer Science

Data Integration and Data Exchange

scoundrel — Thu, 05 Apr 2007 16:19:59 GMT

Google TechTalks
March 24, 2006

Alan Nash

ABSTRACT

I will discuss two fundamental problems in information integration:

how to answer a query over a public interface which combines data from several sources and
how to create a single database conforming to the public interface which combines data from several sources.

I consider the case where the sources are relational databases, where the public interface is a public schema (a specification of the format of a database), and where the sources are related to the public schema by a mapping that is specified by constraints.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, DSP, Computer Science

Detecting and Recognizing Objects In Natural Images

scoundrel — Wed, 04 Apr 2007 11:54:26 GMT

Google Tech Talks
March 9, 2007

ABSTRACT

This talk describes recent work of the UCLA Center for Image and Vision Science for learning to detect and recognize objects in natural images. This includes a real-time system for detecting text in unconstrained environments. Other examples include unsupervised learning of probabilistic grammars for detecting, recognizing, and segmenting objects.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, DSP, Computer Science

Learning Object Appearance Models via Transformed Dirichlet Processes

scoundrel — Thu, 29 Mar 2007 11:17:35 GMT

Google Tech Talks
March 6, 2007

ABSTRACT

Object recognition systems use features extracted from images to localize and categorize objects. Such methods must be robust to the rich variability of natural scenes, and the often small size of training databases. In this talk, we describe hierarchical generative models for objects, the parts composing them, and the scenes surrounding them. We employ Dirichlet processes to learn flexible appearance models which transfer knowledge among related object categories. By coupling part-based models with spatial transformations, we also consistently account for geometric constraints. Through Monte Carlo methods, we use these transformed Dirichlet processes to categorize objects given few examples, and automatically recognize groups of objects in complex visual scenes.

Read more about this video…

Want to control this feed contents? Sign up here and create your own feed!

Want more on these topics?
Browse the archive of posts filed under Conferences, Companies, Broadcasting, Science, Techtalks, Google, Lectures, Math, DSP