Natural Language (NL) descriptions can be one of the most convenient ways to initialize a visual tracker. NL descriptions can also help provide information for longer-term invariance, thus helping the tracker cope better with typical visual tracking challenges, e.g. occlusion, motion blur, etc. However, deriving a formulation to combine the strengths of appearance-based tracking with the NL modality is not straightforward. In this thesis, we use deep neural networks to learn a joint representation of language and vision that can perform various tasks, such as visual tracking by NL, tracked-object retrieval by NL, and spatio-temporal video groundings by NL. First, we study the Single Object Tracking (SOT) by NL descriptions task, which requires spatial localizations of a target object in a video sequence. We propose two novel approaches. The first is a tracking-by-detection approach, which performs object detection in the video sequence via similarity matching between potential objects' pooled visual representations and NL descriptions. The second approach uses a novel Siamese Natural Language Region Proposal Network with a depth-wise cross correlation operation to replace the visual template with a language template in Siamese trackers, e.g. SiamFC, SiamRPN++, etc., and achieves state-of-the-art on standard single object tracking by NL benchmarks. Second, based on experimental results and findings from the SOT by NL task, we propose the Tracked-object Retrieval by NL (TRNL) descriptions task and collect the CityFlow-NL Benchmark for it. CityFlow-NL contains more than 6,500precise NL descriptions of tracked vehicle targets, making it the first densely annotated dataset of tracked-objects paired with NL descriptions. To highlight the novelty of our dataset, we propose two models for the retrieval by NL task: a single-stream model based on cross-modality similarity matching and a quad-stream retrieval model that models the similarity between language features and visual features, including local visual features, frame-level features, motions, and relationships between visually similar targets. We release the CityFlow-NL Benchmark together with our models as challenges in the 5th and the 6th AI City Challenge. Lastly, we focus on the most challenging yet practical task of Spatio-Temporal Video Grounding (STVG), which aims to spatially and temporally localize a target in videos with NL descriptions. We propose new evaluation protocols for the STVG task to adapt to the new challenges of CityFlow-NL that are not well-represented in prior STVG benchmarks. Three intuitive and novel approaches to the STVG task are proposed and studied in this thesis, i.e. Multi-Object Tracking (MOT) + Retrieval by NL approach, Single Object Tracking (SOT) by NL based approach, and a direct localization approach that uses a transformer network to learn a joint representation from both the NL and vision modalities.

The CSPI mini-MAP array processor is supported for use with a RATFOR preprocessor in the software environment defined by the Generalized Image Processing System (GIPSY). A set of interface routines presents the mini-MAP as a tightly-coupled slave processor with well-defined rules for control from the host computer. The slave is programmed by adapting host-based software, using a proscribed set of guidelines for conversion. A software protocol has been defined to allow mini-MAP data memory to be allocated dynamically. Several examples of modified GIPSY commands are examined. / Master of Science

Eidgenössische Techn. Hochsch., Diss.--Zürich, 2003.

Call number: LD2668 .T4 1986 Y84 / Master of Science / Computing and Information Sciences

Typescript (photocopy). / Digitized by Kansas Correctional Industries

Although there is research into student learning on the Web and other computer-supported environments, there has been little investigation into the practice and pedagogy of university Web teachers. This qualitative study used a series of interviews of eight higher education faculty to gather data on their Web teaching practices, and an examination of their Web courses to identify their pedagogies. There was a notable difference in the way the four teachers with Web-assisted courses and the four teachers with Web-only courses used the Web in their teaching. Those with Web-assisted courses used the Web primarily as a connection to expanded resources and a supplement to their face-to-face teaching. Those who taught Web-only courses used the Web for resource access, and also used asynchronous dialogue and peer interactions to support student construction of knowledge. Moreover, the Web-only teachers reported a shift in their roles from lecturer and expert in the classroom to facilitator and co-learner as Web teachers. Four aspects of the Web teaching environment appear to be foundational in supporting an effective Web pedagogy: (1) the varied and extensive uses of e-mail, (2) the "think time" made possible by asynchronous dialogue, (3) distributed, hyperlinked learning, and (4) a reported shift from a content focus to process and issues because Web instructors are confident that the material is presented in the Web course as they wish it. The experiences of the Web teachers in the study would indicate that these four elements can be leveraged to improve university Web teaching and deepen student learning, perhaps even beyond results capable of achievement in face-to-face teaching. / Graduation date: 1999

As the World Wide Web continues to grow, people clearly want to do much more with it than just publish static pages of text and graphics. While such increased interactivity has traditionally been accomplished through the use of server-side CGI scripts, much recent research on Web browsers has been on extending their capabilities through the addition of various types of client-side services. The most popular of these extensions take the form of plug-ins, applets, and "document scripts" such as Java Script. However, because these extensions have been created in a haphazard way by a variety of independent groups, they suffer greatly in terms of flexibility, uniformity, and interoperability. Interactive HTML is a system that addresses these problems by combining plug-ins, applets, and document scripts into one uniform and cohesive architecture. It is implemented as an external C library that can be used by a browser programmer to add client-side services to the browser. The IHTML services are implemented as dynamically loaded "language modules," allowing new plug-ins and language interpreters to be added to an iHTML browser without recompiling the browser itself. The system is currently integrated with NCSA's X Mosaic browser and includes language modules for a text viewer plug-in and Python language interpreter. This thesis examines the iHTML architecture in the context of the historical development of Web client-side services and presents an example of iHTML's use to collect usage information about Web documents. / Graduation date: 1997

Although much effort has been invested to build applications that support group work, collaborative applications have not found easy success. The cost of adopting and maintaining collaborative applications has prevented their widespread use, especially among small distributed groups. Application developers have had difficulties recognizing the extra effort required by groups to use collaborative applications and how to either reduce this effort or provide other benefits to compensate for the extra work. These problems have limited the success of collaborative applications, which have not attained the same level of productivity improvements that single user applications have achieved. In this thesis we present a framework that describes the types of computer support that can facilitate the work of distributed engineering design groups. Our framework addresses support for web-based groups in particular because we believe the web can be a powerful medium for collaboration if accommodated properly. We show how the concepts in this framework can be implemented by prototyping a web-based engineering decision support system. Our framework is a synthesis of ideas motivated by an examination of literature in various fields that share a common interest in collaborative work. It can influence application development by helping developers become aware of the types of support should be considered to aid web-based collaborative design. / Graduation date: 1997

The sustainability and continued development of the chemical industry is to a large extent dependent on learning from past incidents. The failure to learn from past mistakes is rather not deliberate but due to unawareness of the situation. Incident databases are excellent resources to learn from past mistakes; however, in order to be effective, incident databases need to be functional in disseminating the lessons learned to users. Therefore, this research is dedicated to improving user accessibility of incident databases. The objective of this research is twofold. The first objective is improving accessibility of the database system by allowing the option of word search as well as folder search for the users. This will satisfy research needs of users who are aware of the hazards at hand and need to access the relevant information. The second objective is to activate the database via integration of the database with an operational software. This will benefit research needs of users who are unaware of the existing hazards. Literature review and text mining of Major Accident Reporting System (MARS) database short reports are employed to develop an initial taxonomy, which is then refined and modified by expert review. The incident reports in MARS database is classified to the right folders in the taxonomy and implemented in a database system based on Microsoft Excel, where the users can retrieve information using folder search as well as word search option via a user friendly interface. A program coded in JAVA is prepared for integrating the incident database with a Management of Change (MOC) software prototype. A collection of keywords on hazardous substances and equipment is prepared. If the keywords exist in the MOC interface, they will be highlighted, and with the click of a button, will return up to ten relevant incident reports. Using an active and knowledge-based system, people can learn from incidents and near-misses and will be more active to reduce the frequency of recurring incidents.

No description available.