To understand the real-world, it is essential to perceive in all-day conditions including cases which are not suitable for RGB sensors, especially at night. Beyond these limitations, the innovation introduced here is a multispectral solution in the form of depth estimation from a thermal sensor without an additional depth sensor.Based on an analysis of multispectral properties and the relevance to depth predictions, we propose an efficient and novel multi-task framework called the Multispectral Transfer Network (MTN) to estimate a depth image from a single thermal image. By exploiting geometric priors and chromaticity clues, our model can generate a pixel-wise depth image in an unsupervised manner. Moreover, we propose a new type of multitask module called Interleaver as a means of incorporating the chromaticity and fine details of skip-connections into the depth estimation framework without sharing feature layers. Lastly, we explain a novel technical means of stably training and covering large disparities and extending thermal images to data-driven methods for all-day conditions. In experiments, we demonstrate the better performance and generalization of depth estimation through the proposed multispectral stereo dataset, including various driving conditions.

The fashion industry is establishing its presence on a number of visual-centric social media like Instagram. This creates an interesting clash as fashion brands that have traditionally practiced highly creative and editorialized image marketing now have to engage with people on the platform that epitomizes impromptu, realtime conversation. What kinds of fashion images do brands and individuals share and what are the types of visual features that attract likes and comments? In this research, we take both quantitative and qualitative approaches to answer these questions. We analyze visual features of fashion posts first via manual tagging and then via training on convolutional neural networks. The classified images were examined across four types of fashion brands: mega couture, small couture, designers, and high street. We find that while product-only images make up the majority of fashion conversation in terms of volume, body snaps and face images that portray fashion items more naturally tend to receive a larger number of likes and comments by the audience. Our findings bring insights into building an automated tool for classifying or generating influential fashion information. We make our novel dataset of 24,752 labeled images on fashion conversations, containing visual and textual cues, available for the research community.

Public discourse on environmental and health issues has risenon social media. Upon an environmental crisis, various chatterssuch as breaking news, misinformation, and rumor couldaggravate social confusion and proliferate negative publicsentiment. In an effort to study public sentiments on environmentalissues in South Korea, we analyzed 158,964 tweetsgenerated over a 4-year period following the Fukushima accidentin 2011, the largest release of radioactivity to environmentin recent history. This event led to a significant increasein public’s interest on environmental and nuclear issues inKorea. We employed Bayesian network and recursive partitioningto observe the classification regression tree structureof major topics. Topics on health and environment were interlinkedclosely and represented both apprehension and concernabout health threats and pollution. Our methodologyhelps analyze large online discourse efficiently and offers insightto crisis response organizations.

For many people who speak more than one language,their language proficiency for each of the languagesvaries. We can conjecture that people who use onelanguage (primary language) more than another wouldshow higher language proficiency in that primary language.It is, however, difficult to observe and quantifythat problem because natural language use is difficultto collect in large amounts. We identify Wikipedia asa great resource for studying multilingualism, and weconduct a quantitative analysis of the language complexityof primary and non-primary users of English,German, and Spanish. Our preliminary results indicatethat there are indeed consistent differences of languagecomplexity in the Wikipedia articles chosen by primaryand non-primary users, as well as differences in the editsby the two groups of users.

We propose a new customizable tool, Language Independent Feature Extractor (LIFE), which models the inherent patterns of any language and extracts relevant features of thelanguage. There are two contributions of this work: (1) no labeled data is necessary to train LIFE (It works when a sufficient number of unlabeled documents are given), and (2) LIFE is designed to be applicable to any language. We proved the usefulness of LIFE by experimental results of time information extraction.

In this paper, we build a social search engine named Glaucus for location-based queries. They compose a significant portion of mobile searches, thus becoming more popular with the prevalence of mobile devices. However, most of existing social search engines are not designed for location-based queries and thus often produce poor-quality results for such queries. Glaucus is inherently designed to support location-based queries. It collects the check-in information, which pinpoints the places where each user visited, from location-based social networking services such as Foursquare. Then, it calculates the expertise of each user for a query by using our new probabilistic model called the location aspect model . We conducted two types of evaluation to prove the effectiveness of our engine. The results showed that Glaucus selected the users supported by stronger evidence for the required expertise than existing social search engines. In addition, the answers from the experts selected by Glaucus were highly rated by our human judges in terms of answer satisfaction.

Community-based question answering (CQA) services such as Yahoo! Answers have been widely used by Internet users to get the answers for their inquiries. The CQA services totally rely on the contributions by the users. However, it is known that newcomers are prone to lose their interests and leave the communities. Thus, finding expert users in an early phase when they are still active is essential to improve the chances of motivating them to contribute to the communities further. In this paper, we propose a novel approach to discovering "potentially" contributive users from recently-joined users in CQA services. The likelihood of becoming a contributive user is defined by the user's expertise as well as availability, which we call the answer affordance. The main technical difficulty lies in the fact that such recently-joined users do not have abundant information accumulated for many years. We utilize a user's productive vocabulary to mitigate the lack of available information since the vocabulary is the most fundamental element that reveals his/her knowledge. Extensive experiments were conducted with a huge data set of Naver Knowledge-In (KiN), which is the dominating CQA service in Korea. We demonstrate that the top rankers selected by the answer affordance outperformed those by KiN in terms of the amount of answering activity.