Disaster Monitoring using Unmanned Aerial Vehicles and Deep LearningAndreas Kamilaris
Monitoring and identification of disasters are crucial for mitigating their effects on the
environment and on human population, and can be facilitated by the use of unmanned aerial vehicles
(UAV), equipped with camera sensors which can produce frequent aerial photos of the areas of interest. A
modern, promising technique for recognition of events based on aerial photos is deep learning. In this paper,
we present the state of the art work related to the use of deep learning techniques for disaster monitoring
and identification. Moreover, we demonstrate the potential of this technique in identifying disasters
automatically, with high accuracy, by means of a relatively simple deep learning model. Based on a small
dataset of 544 images (containing images of disasters such as fires, earthquakes, collapsed buildings,
tsunami and flooding, as well as “non-disaster” scenes), our preliminary results show an accuracy of 91%
achieved, indicating that deep learning, combined with UAV equipped with camera sensors, have the
potential to predict disasters with high accuracy in the near future. Presented at the EnviroInfo 2017 Conference in Luxembourg.
Troyes University of Technology (UTT) is a public engineering university located in Troyes, France. It has 2,600 students, 175 PhD students, and offers 6 engineering degree programs. UTT has a three-fold mission of doctoral education, undergraduate degrees, and continuing education. It has strong international partnerships, customized degree plans, and focuses on giving students real-world experience through internships. UTT's research is conducted through 8 teams affiliated with the Charles Delaunay Institute and focuses on areas like risk management and sustainable development. The university has over 3,000 business partners and generates revenue through private research contracts.
Articial societies immersed in an Ambient Intelligence EnvironmentEmilio Serrano
1) The document proposes using multi-agent based simulations (MABS) to develop and test ambient intelligence (AmI) applications for emergencies, as real tests can be too costly or impractical.
2) It describes using MABS to simulate an emergency evacuation of a real university building, including a building model, people, fire, and AmI devices like sensors and actuators.
3) Three example AmI applications for evacuation are tested - one that deactivates devices, one that activates all on fire, and one that also provides evacuation directions. Simulations allow analyzing many parameters like evacuation times.
This document summarizes CBIR with LIRe, an open source Java-based image retrieval engine based on Lucene. It provides an overview of LIRe's features such as color histograms, MPEG-7 descriptors, and local features. It also discusses LIRe's indexing and search capabilities, examples of its use, evaluation in image retrieval benchmarks, and potential future directions.
LIRE presentation at the ACM Multimedia Open Source Software Competition 2013dermotte
This document summarizes the LIRE (Library for Content Based Image Retrieval) framework. It describes LIRE as a Java-based CBIR library built on Lucene that provides easy indexing and searching of image features. Key points include that it has a modular feature architecture, supports fast linear and sub-linear search through hashing techniques, and includes tools for parallel indexing and intermediate data formats. The document also briefly discusses integrating LIRE with Apache Solr and future work plans.
Disaster Monitoring using Unmanned Aerial Vehicles and Deep LearningAndreas Kamilaris
Monitoring and identification of disasters are crucial for mitigating their effects on the
environment and on human population, and can be facilitated by the use of unmanned aerial vehicles
(UAV), equipped with camera sensors which can produce frequent aerial photos of the areas of interest. A
modern, promising technique for recognition of events based on aerial photos is deep learning. In this paper,
we present the state of the art work related to the use of deep learning techniques for disaster monitoring
and identification. Moreover, we demonstrate the potential of this technique in identifying disasters
automatically, with high accuracy, by means of a relatively simple deep learning model. Based on a small
dataset of 544 images (containing images of disasters such as fires, earthquakes, collapsed buildings,
tsunami and flooding, as well as “non-disaster” scenes), our preliminary results show an accuracy of 91%
achieved, indicating that deep learning, combined with UAV equipped with camera sensors, have the
potential to predict disasters with high accuracy in the near future. Presented at the EnviroInfo 2017 Conference in Luxembourg.
Troyes University of Technology (UTT) is a public engineering university located in Troyes, France. It has 2,600 students, 175 PhD students, and offers 6 engineering degree programs. UTT has a three-fold mission of doctoral education, undergraduate degrees, and continuing education. It has strong international partnerships, customized degree plans, and focuses on giving students real-world experience through internships. UTT's research is conducted through 8 teams affiliated with the Charles Delaunay Institute and focuses on areas like risk management and sustainable development. The university has over 3,000 business partners and generates revenue through private research contracts.
Articial societies immersed in an Ambient Intelligence EnvironmentEmilio Serrano
1) The document proposes using multi-agent based simulations (MABS) to develop and test ambient intelligence (AmI) applications for emergencies, as real tests can be too costly or impractical.
2) It describes using MABS to simulate an emergency evacuation of a real university building, including a building model, people, fire, and AmI devices like sensors and actuators.
3) Three example AmI applications for evacuation are tested - one that deactivates devices, one that activates all on fire, and one that also provides evacuation directions. Simulations allow analyzing many parameters like evacuation times.
This document summarizes CBIR with LIRe, an open source Java-based image retrieval engine based on Lucene. It provides an overview of LIRe's features such as color histograms, MPEG-7 descriptors, and local features. It also discusses LIRe's indexing and search capabilities, examples of its use, evaluation in image retrieval benchmarks, and potential future directions.
LIRE presentation at the ACM Multimedia Open Source Software Competition 2013dermotte
This document summarizes the LIRE (Library for Content Based Image Retrieval) framework. It describes LIRE as a Java-based CBIR library built on Lucene that provides easy indexing and searching of image features. Key points include that it has a modular feature architecture, supports fast linear and sub-linear search through hashing techniques, and includes tools for parallel indexing and intermediate data formats. The document also briefly discusses integrating LIRE with Apache Solr and future work plans.
This document provides an overview of visual information retrieval and content-based image retrieval. It discusses the motivation for these topics due to the large number of digital images and videos now available. Various low-level image features that can be extracted for content-based image retrieval are described, including color histograms, dominant colors, color distribution, and color correlograms. The benefits and disadvantages of different features are also outlined. The goal is to reduce images to feature vectors for efficient similarity comparison and retrieval.
The document discusses the European Commission's efforts to support e-infrastructure and e-science. It outlines several key initiatives, including the Digital Agenda for Europe, which aims to maximize benefits of digital technologies. Major projects include the PRACE partnership for advanced supercomputing and efforts to build pan-European high-speed networks like GÉANT to enable scientific collaboration and access to data resources across borders. The Commission sees e-infrastructure as crucial to enabling cutting-edge e-science and helping European researchers address global challenges.
Artificial intelligence in cyber defenseDinesh More
This document discusses the use of artificial intelligence in cyber defense. It argues that AI is needed to rapidly analyze large amounts of data and react to the evolving threats in cyberspace, which humans cannot do alone without considerable automation. The document provides an overview of artificial intelligence as a field, and surveys existing AI applications in cyber defense that use techniques like artificial neural networks, expert systems, intelligent agents, search algorithms, machine learning, and data mining. It concludes that while useful AI applications already exist, many cyber defense problems could be better solved by incorporating more intelligent methods.
Int. Workshop on Information Systems for Social Innovation (ISSI) 2013 Session: Systems Resilience
National Institute of Informatics, Tokyo, Japan
February 4, 2014
http://tric.rois.ac.jp/human/ISSI2013/
The document discusses computer vision and its history. Computer vision involves using algorithms to understand and analyze visual images and video data. It aims to help computers understand scenes, locate objects, and determine their properties similarly to human vision. Computer vision has many applications such as face detection and recognition, optical character recognition, analyzing sports footage, and enabling technologies like autonomous vehicles and robots. The field involves understanding problems like image formation, filtering, matching, alignment, and categorization. OpenCV is also introduced as a popular open-source library for computer vision applications.
This document discusses power laws, Zipf's law, and Pareto distributions. It explains that these distributions emerge when human behavior and decisions are involved. While the theories behind these distributions are well-established, there are also practical questions around how to detect and utilize them, such as how to identify content that may become popular and optimize websites, videos, and other resources accordingly. The document concludes that these distributions are real phenomena but the exact mechanisms behind why they occur are still not fully understood.
Virtual reality (VR) allows users to interact with simulated 3D environments through specialized computer hardware and software. This document describes the process of creating VR simulations of the human skull and heart using medical imaging data. CT and MRI scans were converted into 3D models and rendered into stereo image sequences. These sequences could be viewed through VR headsets or monitors to provide an interactive educational experience, allowing students to fly through virtual tours of the human body. While high resolution VR is possible, limitations remain in hardware and software. Researchers continue working to improve VR technology for medical education applications.
Bridging the Gap: Machine Learning for Ubiquitous Computing -- IntroductionThomas Ploetz
Tutorial @Ubicomp 2015: Bridging the Gap -- Machine Learning for Ubiquitous Computing (introduction session).
A tutorial on promises and pitfalls of Machine Learning for Ubicomp (and Human Computer Interaction). From Practitioners for Practitioners.
Presenter: Thomas Ploetz <tom.ploetz@gmail.com>
video recording of talks as they wer held at Ubicomp:
https://youtu.be/LgnnlqOIXJc?list=PLh96aGaacSgXw0MyktFqmgijLHN-aQvdq
This document summarizes a machine learning meetup in Sofia. It discusses trends in cognitive computing and machine learning, including computers that learn, think, interact with humans and other computers. It also outlines enabling technologies for cognitive computing like natural language processing. Specific machine learning tasks like classification, regression and clustering are covered. Challenges in machine learning like data requirements and training time are addressed. The document promotes sharing knowledge and ideas at the open meetup format.
Opportunities and Challenges of Large-scale IoT Data AnalyticsPayamBarnaghi
The document discusses opportunities and challenges of large-scale IoT data analytics. It provides an overview of the evolution of IoT from early technologies to current applications and future directions. It describes the types of heterogeneous and real-time data generated by IoT devices and challenges in analyzing this data. Examples of applications discussed include smart cities, transportation, healthcare, and event analysis. The document also summarizes work done in the EU CityPulse project on extracting events from social media and demonstrating IoT data analytics techniques.
For the full video of this presentation, please visit:
http://www.embedded-vision.com/platinum-members/embedded-vision-alliance/embedded-vision-training/videos/pages/dec-2016-member-meeting-khronos
For more information about embedded vision, please visit:
http://www.embedded-vision.com
Mark Bünger, Vice President of Research at Lux Research, delivers the presentation "Imaging + AI: Opportunities Inside the Car and Beyond" at the December 2016 Embedded Vision Alliance Member Meeting. Bünger presents his firm’s perspective on how embedded vision will upend the automotive industry.
Archaeology & cultural heritage application working group part 2Manolis Vavalis
The document summarizes the proceedings of a review meeting on archaeology and cultural heritage applications. It lists the members of the application working group from the first and second years. It then provides details on a thematic workshop organized by the group on 3D knowledge technologies, including the program, position statements discussed, and outcomes. It also describes scenarios for virtual exhibitions, integrating geometry and knowledge, and animating virtual human crowds. Open problems addressed include facilitating automatic semantic annotation of 3D content and enhancing repositories to exploit semantics.
Google Trends and other IT fever charts rate Data
Science among the most rapidly emerging and promising fields that expand around computer science. Although Data Science draws on content from established fields like artificial intelligence, statistics, databases, visualization and many more, industry is demanding for trained data scientists that no one seems able to deliver. This is due to the pace at which the field has expanded and the corresponding lack of curricula; the
unique skill set, which is inherently multi-disciplinary; and the translation work (from the US web economy to other ecosystems) necessary to realize the recognized world-wide potential of applying analytics to all sorts of data.
In this contribution we draw from our experiences in establishing an inter-disciplinary Data Science lab in order to highlight the challenges and potential remedies for Data Science
in Europe. We discuss our role as academia in the light of the potential societal/economic impact as well as the challenges in organizational leadership tied to such inter-disciplinary work.
Penn State researchers used 3D printing technology to create realistic busts of two unidentified children based on skeletal remains. A senior project associate printed the busts from 3D image files created through a haptic interface, allowing for identification of physical features. The busts have helped identify one child as a boy who died in 1852. This new application of 3D printing could provide a way to help identify missing children similar to current uses of photos and DNA.
Currency Recognition System for Visually Impaired: Egyptian Banknote as a Stu...DrNoura Semary
The document describes a currency recognition system that uses image processing and template matching to identify Egyptian paper money denominations. A camera captures an image of a bill which undergoes preprocessing like thresholding and noise removal before extracting the region of interest. Template matching is then used to compare the region of interest to templates of each denomination in a dataset, identifying the bill based on the highest correlation. The system achieved an average accuracy of 89% across 120 test images of various denominations. Future work aims to improve accuracy with feature-based models and incorporate additional currencies.
Internet of Things and Large-scale Data Analytics PayamBarnaghi
This document discusses Internet of Things (IoT) and large-scale data analytics. It begins by noting the increasing capabilities of computing devices over time, from early mainframes to modern smartphones. It then discusses the growing number of connected sensors, devices, and "things" that are part of the IoT. The document outlines some of the challenges around IoT and big data, such as heterogeneous, noisy data from many sources. It presents examples of applying IoT and analytics to problems in smart cities. Specifically, it discusses using sensor data for applications like transportation optimization and power grid management. The conclusion emphasizes that IoT analytics requires approaches that can handle resource constraints and cross-layer optimizations across the network architecture.
This document discusses a lecture on computer vision given by Dr. Eng. Mahmoud Shams at Kafrelsheikh University. It defines computer vision as dealing with how computers understand digital images and videos, and seeks to automate tasks of the human visual system. The lecture covers classification of AI, evaluation of computer vision algorithms, common computer vision tasks like localization and segmentation, and why benchmarks are important. It also lists the top 10 computer vision tools for 2020 and discusses negative results in computer vision research.
This document discusses a lecture on computer vision given by Dr. Eng. Mahmoud Shams at Kafrelsheikh University. It defines computer vision as dealing with how computers understand digital images and videos, and seeks to automate tasks of the human visual system. The lecture covers classification of AI, evaluation of computer vision algorithms, common computer vision tasks like localization and segmentation, and why benchmarks are important. It also discusses sources of noise in images, performance metrics like mean square error and confusion matrices, and some top computer vision tools like OpenCV, TensorFlow, Keras and YOLO.
This document discusses artificial intelligence (AI) and provides several quotes about AI from experts such as Stephen Hawking, Ray Kurzweil, Elon Musk, and others. It then summarizes the history of AI and key developments that led to the current "third AI boom". These include advances in machine learning, deep learning, self-driving cars, smart assistants, and more. The document also discusses challenges for AI such as the need for AI systems to interact and react, as well as the impact of AI on jobs and the need for reskilling workers.
European Data Science Academy - Enabling Data Driven Digital EuropePersontyle
The ‘Age of Data’ continues to thrive, with data being produced from all industries at a phenomenal rate that introduces numerous challenges regarding the collection, storage and analysis of this data. To address this problem, the European Data Science Academy (EDSA) will establish a virtuous learning production cycle for Data Science.
To learn more about the project visit: http://edsa-project.eu/
Invited Talk OAGM Workshop Salzburg, May 2015dermotte
There is a gap between a user's information need and the queries they submit, known as the "intention gap". Bridging this gap is challenging due to the difficulty of translating intentions into search queries. Researchers have studied user intentions in various contexts like search, media production and sharing. However, fully understanding intentions is difficult as people have trouble expressing their own intentions and judging those of others. Future work should develop new techniques to relate content-based image retrieval to user intentions and take an interdisciplinary approach to better model intentions across domains.
CBMI 2013 Presentation: User Intentions in Multimediadermotte
This document discusses user intentions in visual information retrieval and multimedia information systems. It begins by introducing query by example search and different low-level visual features that work better for some domains than others. It then discusses how determining the right features and defining visual similarity is challenging. The document defines context and intention, and discusses how a user's intention relates to their information need. It reviews taxonomies of user intentions in web search and proposes intentions in multimedia may include search, production, sharing, archiving. The document proposes several open PhD theses around developing a general model of user intentions in multimedia, using games and human computation to infer intentions, bringing context to queries, and creating adaptable applications based on user intentions.
This document provides an overview of visual information retrieval and content-based image retrieval. It discusses the motivation for these topics due to the large number of digital images and videos now available. Various low-level image features that can be extracted for content-based image retrieval are described, including color histograms, dominant colors, color distribution, and color correlograms. The benefits and disadvantages of different features are also outlined. The goal is to reduce images to feature vectors for efficient similarity comparison and retrieval.
The document discusses the European Commission's efforts to support e-infrastructure and e-science. It outlines several key initiatives, including the Digital Agenda for Europe, which aims to maximize benefits of digital technologies. Major projects include the PRACE partnership for advanced supercomputing and efforts to build pan-European high-speed networks like GÉANT to enable scientific collaboration and access to data resources across borders. The Commission sees e-infrastructure as crucial to enabling cutting-edge e-science and helping European researchers address global challenges.
Artificial intelligence in cyber defenseDinesh More
This document discusses the use of artificial intelligence in cyber defense. It argues that AI is needed to rapidly analyze large amounts of data and react to the evolving threats in cyberspace, which humans cannot do alone without considerable automation. The document provides an overview of artificial intelligence as a field, and surveys existing AI applications in cyber defense that use techniques like artificial neural networks, expert systems, intelligent agents, search algorithms, machine learning, and data mining. It concludes that while useful AI applications already exist, many cyber defense problems could be better solved by incorporating more intelligent methods.
Int. Workshop on Information Systems for Social Innovation (ISSI) 2013 Session: Systems Resilience
National Institute of Informatics, Tokyo, Japan
February 4, 2014
http://tric.rois.ac.jp/human/ISSI2013/
The document discusses computer vision and its history. Computer vision involves using algorithms to understand and analyze visual images and video data. It aims to help computers understand scenes, locate objects, and determine their properties similarly to human vision. Computer vision has many applications such as face detection and recognition, optical character recognition, analyzing sports footage, and enabling technologies like autonomous vehicles and robots. The field involves understanding problems like image formation, filtering, matching, alignment, and categorization. OpenCV is also introduced as a popular open-source library for computer vision applications.
This document discusses power laws, Zipf's law, and Pareto distributions. It explains that these distributions emerge when human behavior and decisions are involved. While the theories behind these distributions are well-established, there are also practical questions around how to detect and utilize them, such as how to identify content that may become popular and optimize websites, videos, and other resources accordingly. The document concludes that these distributions are real phenomena but the exact mechanisms behind why they occur are still not fully understood.
Virtual reality (VR) allows users to interact with simulated 3D environments through specialized computer hardware and software. This document describes the process of creating VR simulations of the human skull and heart using medical imaging data. CT and MRI scans were converted into 3D models and rendered into stereo image sequences. These sequences could be viewed through VR headsets or monitors to provide an interactive educational experience, allowing students to fly through virtual tours of the human body. While high resolution VR is possible, limitations remain in hardware and software. Researchers continue working to improve VR technology for medical education applications.
Bridging the Gap: Machine Learning for Ubiquitous Computing -- IntroductionThomas Ploetz
Tutorial @Ubicomp 2015: Bridging the Gap -- Machine Learning for Ubiquitous Computing (introduction session).
A tutorial on promises and pitfalls of Machine Learning for Ubicomp (and Human Computer Interaction). From Practitioners for Practitioners.
Presenter: Thomas Ploetz <tom.ploetz@gmail.com>
video recording of talks as they wer held at Ubicomp:
https://youtu.be/LgnnlqOIXJc?list=PLh96aGaacSgXw0MyktFqmgijLHN-aQvdq
This document summarizes a machine learning meetup in Sofia. It discusses trends in cognitive computing and machine learning, including computers that learn, think, interact with humans and other computers. It also outlines enabling technologies for cognitive computing like natural language processing. Specific machine learning tasks like classification, regression and clustering are covered. Challenges in machine learning like data requirements and training time are addressed. The document promotes sharing knowledge and ideas at the open meetup format.
Opportunities and Challenges of Large-scale IoT Data AnalyticsPayamBarnaghi
The document discusses opportunities and challenges of large-scale IoT data analytics. It provides an overview of the evolution of IoT from early technologies to current applications and future directions. It describes the types of heterogeneous and real-time data generated by IoT devices and challenges in analyzing this data. Examples of applications discussed include smart cities, transportation, healthcare, and event analysis. The document also summarizes work done in the EU CityPulse project on extracting events from social media and demonstrating IoT data analytics techniques.
For the full video of this presentation, please visit:
http://www.embedded-vision.com/platinum-members/embedded-vision-alliance/embedded-vision-training/videos/pages/dec-2016-member-meeting-khronos
For more information about embedded vision, please visit:
http://www.embedded-vision.com
Mark Bünger, Vice President of Research at Lux Research, delivers the presentation "Imaging + AI: Opportunities Inside the Car and Beyond" at the December 2016 Embedded Vision Alliance Member Meeting. Bünger presents his firm’s perspective on how embedded vision will upend the automotive industry.
Archaeology & cultural heritage application working group part 2Manolis Vavalis
The document summarizes the proceedings of a review meeting on archaeology and cultural heritage applications. It lists the members of the application working group from the first and second years. It then provides details on a thematic workshop organized by the group on 3D knowledge technologies, including the program, position statements discussed, and outcomes. It also describes scenarios for virtual exhibitions, integrating geometry and knowledge, and animating virtual human crowds. Open problems addressed include facilitating automatic semantic annotation of 3D content and enhancing repositories to exploit semantics.
Google Trends and other IT fever charts rate Data
Science among the most rapidly emerging and promising fields that expand around computer science. Although Data Science draws on content from established fields like artificial intelligence, statistics, databases, visualization and many more, industry is demanding for trained data scientists that no one seems able to deliver. This is due to the pace at which the field has expanded and the corresponding lack of curricula; the
unique skill set, which is inherently multi-disciplinary; and the translation work (from the US web economy to other ecosystems) necessary to realize the recognized world-wide potential of applying analytics to all sorts of data.
In this contribution we draw from our experiences in establishing an inter-disciplinary Data Science lab in order to highlight the challenges and potential remedies for Data Science
in Europe. We discuss our role as academia in the light of the potential societal/economic impact as well as the challenges in organizational leadership tied to such inter-disciplinary work.
Penn State researchers used 3D printing technology to create realistic busts of two unidentified children based on skeletal remains. A senior project associate printed the busts from 3D image files created through a haptic interface, allowing for identification of physical features. The busts have helped identify one child as a boy who died in 1852. This new application of 3D printing could provide a way to help identify missing children similar to current uses of photos and DNA.
Currency Recognition System for Visually Impaired: Egyptian Banknote as a Stu...DrNoura Semary
The document describes a currency recognition system that uses image processing and template matching to identify Egyptian paper money denominations. A camera captures an image of a bill which undergoes preprocessing like thresholding and noise removal before extracting the region of interest. Template matching is then used to compare the region of interest to templates of each denomination in a dataset, identifying the bill based on the highest correlation. The system achieved an average accuracy of 89% across 120 test images of various denominations. Future work aims to improve accuracy with feature-based models and incorporate additional currencies.
Internet of Things and Large-scale Data Analytics PayamBarnaghi
This document discusses Internet of Things (IoT) and large-scale data analytics. It begins by noting the increasing capabilities of computing devices over time, from early mainframes to modern smartphones. It then discusses the growing number of connected sensors, devices, and "things" that are part of the IoT. The document outlines some of the challenges around IoT and big data, such as heterogeneous, noisy data from many sources. It presents examples of applying IoT and analytics to problems in smart cities. Specifically, it discusses using sensor data for applications like transportation optimization and power grid management. The conclusion emphasizes that IoT analytics requires approaches that can handle resource constraints and cross-layer optimizations across the network architecture.
This document discusses a lecture on computer vision given by Dr. Eng. Mahmoud Shams at Kafrelsheikh University. It defines computer vision as dealing with how computers understand digital images and videos, and seeks to automate tasks of the human visual system. The lecture covers classification of AI, evaluation of computer vision algorithms, common computer vision tasks like localization and segmentation, and why benchmarks are important. It also lists the top 10 computer vision tools for 2020 and discusses negative results in computer vision research.
This document discusses a lecture on computer vision given by Dr. Eng. Mahmoud Shams at Kafrelsheikh University. It defines computer vision as dealing with how computers understand digital images and videos, and seeks to automate tasks of the human visual system. The lecture covers classification of AI, evaluation of computer vision algorithms, common computer vision tasks like localization and segmentation, and why benchmarks are important. It also discusses sources of noise in images, performance metrics like mean square error and confusion matrices, and some top computer vision tools like OpenCV, TensorFlow, Keras and YOLO.
This document discusses artificial intelligence (AI) and provides several quotes about AI from experts such as Stephen Hawking, Ray Kurzweil, Elon Musk, and others. It then summarizes the history of AI and key developments that led to the current "third AI boom". These include advances in machine learning, deep learning, self-driving cars, smart assistants, and more. The document also discusses challenges for AI such as the need for AI systems to interact and react, as well as the impact of AI on jobs and the need for reskilling workers.
European Data Science Academy - Enabling Data Driven Digital EuropePersontyle
The ‘Age of Data’ continues to thrive, with data being produced from all industries at a phenomenal rate that introduces numerous challenges regarding the collection, storage and analysis of this data. To address this problem, the European Data Science Academy (EDSA) will establish a virtuous learning production cycle for Data Science.
To learn more about the project visit: http://edsa-project.eu/
Invited Talk OAGM Workshop Salzburg, May 2015dermotte
There is a gap between a user's information need and the queries they submit, known as the "intention gap". Bridging this gap is challenging due to the difficulty of translating intentions into search queries. Researchers have studied user intentions in various contexts like search, media production and sharing. However, fully understanding intentions is difficult as people have trouble expressing their own intentions and judging those of others. Future work should develop new techniques to relate content-based image retrieval to user intentions and take an interdisciplinary approach to better model intentions across domains.
CBMI 2013 Presentation: User Intentions in Multimediadermotte
This document discusses user intentions in visual information retrieval and multimedia information systems. It begins by introducing query by example search and different low-level visual features that work better for some domains than others. It then discusses how determining the right features and defining visual similarity is challenging. The document defines context and intention, and discusses how a user's intention relates to their information need. It reviews taxonomies of user intentions in web search and proposes intentions in multimedia may include search, production, sharing, archiving. The document proposes several open PhD theses around developing a general model of user intentions in multimedia, using games and human computation to infer intentions, bringing context to queries, and creating adaptable applications based on user intentions.
This document summarizes an exploratory study on user intentions for video production. The study aimed to determine if a taxonomy could be developed to classify intentions, and to test current approaches. 20 participants were interviewed about 48 video recording situations. The situations were clustered into categories including preservation, sharing, affection, functional, and technical interest. Nearly all videos were intended for sharing, and most instances fit into multiple categories, suggesting current taxonomies are insufficient and classes are not disjoint for video production intentions. Future work is needed using larger and more varied datasets.
Callisto: Content Based Tag Recommendation for Imagesdermotte
Callisto is a tool that provides ranked tag recommendations for images based on their visual content and any start tags provided. It uses two models: 1) NCP, which bases suggestions on visual features in the image, and 2) a statistical model that analyzes tag co-occurrence. NCP often suggests tags highly related to visual features present in an image and does not suggest tags for features missing from the image. Callisto demonstrates this behavior through examples, showing it can provide different, visually-relevant suggestions for the same start tag depending on the image's actual visual content.
User Intentions or "The other end of the camera ..."dermotte
The document discusses user intentions and goals in interacting with multimedia and computer systems. It proposes that understanding user intentions can help systems better support users. It reviews research on classifying user goals in web search and digital photo retrieval. Studies found goals are difficult to classify as they can be implicit and transition between goals is fuzzy. The document also discusses using intention-based annotation tools for media production. Overall, it argues that making user intentions explicit could help bridge semantic gaps and enhance content retrieval and visualization in multimedia information management systems.
Using Visual Features to Improve Tag Suggestions in Image Sharing Sites :: pr...dermotte
The document discusses using visual features to improve tag suggestions for images on sharing sites. It presents the motivation for improving tagging rates, as only 20-25% of images are currently tagged on sites like Flickr despite the obvious benefits. The authors propose an architecture to focus on the annotation process by using visual features to reduce the effort required for users to tag images. They provide an overview of the current state of their work and preliminary conclusions.
This document analyzes aspects of broad folksonomies based on a sample of over 800,000 bookmarks from Delicious. It finds that:
1) The frequency-rank distribution of co-occurring tags follows a power law for around 80% of tags analyzed.
2) When analyzing resource-based and user-based tagging characteristics, only around 18-13% of tags respectively followed a power law distribution, indicating folksonomies have a more complex underlying structure.
3) Tags provide added value for retrieval over title alone, though precision and recall were mostly below 0.5, showing tags are not redundant but current usage may not optimize retrieval performance.
4. ITEC, Klagenfurt University, Austria Number of Digital Photos(global) Estimate 2006 > 150 billion photos from cameras > 100 billion photos from camera phones Forecast 2010 > 500 billion photos + increased resolution Source: IDC Study “Expanding Digital Universe” http://www.emc.com/about/destination/digital_universe/
5. ITEC, Klagenfurt University, Austria Digital Imaging Devices(Germany) Still image cameras sold in Germany (thousands) analogue digital Source: Cewe Factbook, http://www.cewecolor.de
15. What is the problem with VIR? The fundamental difficulty in doing what we want to do is related to the need to encode, perceive, convey, and measure similarity (e.g. between two images)
16. Bildsuche Vergleich von 2 MP Bildern á 1.600 x 1.200 x 2 byte = 3,66 MB Annahme: 7000 Bilder ~ 25 GB Daten (CC) JasonRogers: http://flickr.com/photos/restlessglobetrotter/2149696743/