We propose a lightweight and highly accurate method for detecting anomalies in videos. Existing methods use multiple-instance learning (MIL) to determine the normal/abnormal status of each segment of the video. Recent successful researches argue that it is important to learn the temporal relationships among segments to achieve high accuracy, instead of focusing on only a single segment. We analyzed the existing methods that have been successful in recent years, and found that while it is indeed important to learn all segments together, the temporal relationships among them are irrelevant to achieving high accuracy. Based on this finding, we do not use the MIL framework, but instead introduce a self-attention mechanism to automatically extract features that are important for determining normal/abnormal from all input segments. As a result, the neural network with 1.3% of the number of parameters of the existing method can achieve the comparable or better accuracy than the existing method.

In this paper, we propose a modeling method of clouds from a single photograph. To synthesize realistic images of outdoor scenes, modeling of 3D clouds is necessary. However, synthesizing realistic clouds is still a difficult task. To address this problem, we propose an example-based volume texture synthesis to model nonhomogeneous density volumes. Our method prepares an example cloud volume data using a physically-based fluid simulator. Then the output cloud volume is automatically synthesized from the example cloud volume data so that its rendered image becomes similar to the clouds in the input photograph. In contrast to the previous texture synthesis methods, our method synthesizes density volumes taking into account scattering and transmittance of light. We show several cloud volumes whose rendered images are similar to the input photograph.

We propose a method of three-dimensional (3D) modeling of volumetric fluid phenomena from sparse multi-view images (e.g., only a single-view input or a pair of front- and side-view inputs). The volume determined from such sparse inputs using previous methods appears blurry and unnatural with novel views; however, our method preserves the appearance of novel viewing angles by transferring the appearance information from input images to novel viewing angles. For appearance information, we use histograms of image intensities and steerable coefficients. We formulate the volume modeling as an energy minimization problem with statistical hard constraints, which is solved using an expectation maximization (EM)-like iterative algorithm. Our algorithm begins with a rough estimate of the initial volume modeled from the input images, followed by an iterative process whereby we first render the images of the current volume with novel viewing angles. Then, we modify the rendered images by transferring the appearance information from the input images, and we thereafter model the improved volume based on the modified images. We iterate these operations until the volume converges. We demonstrate our method successfully provides natural-looking volume sequences of fluids (i.e., fire, smoke, explosions, and a water splash) from sparse multi-view videos. To create production-ready fluid animations, we further propose a method of rendering and editing fluids using a commercially available fluid simulator.

We present KikuNavi, a real-time pedestrian navigation system as a social networking service operating on a mobile phone. A user who becomes lost can use our system to easily find someone (the navigator) who knows how to reach the user's destination. The navigator can sketch the route, which appears on the lost user's mobile phone. The navigator can also choose to lead the user; the lost user's phone displays an arrow pointing toward the navigator. Most existing navigation systems rely on digital maps and GPS geospatial information, but our system is based on the fact that we often become lost in situations when the combination of digital maps and GPS does not work well. Our concept is to support pedestrian navigation even in such an environment by relying on the collective intelligence of people nearby, connected by mobile phones.

We propose a novel method to improve the prediction accuracy on the rating prediction task by correcting the bias of user ratings. We demonstrate that the manner of user rating and review is biased and that it is necessary to correct this difference for more accurate prediction. Our proposed method comprises approaches based on the detection of each user value to ratings: The bias of the rating is detected using entropy of user rating and by updating word weights only when the words appear in the review, the problem of bias is reduced. We implement this idea by extending the Prank algorithm. We apply a review -- item matrix as a feature matrix instead of a user -- item matrix because of its volume of information. Our quantitative evaluation shows that our method improves the prediction accuracy (the Rank Loss measurement) significantly by 8.70 % compared with the normal Prank algorithm. Our proposed method helps users find out what they care about when buying something, and is applicable to newer variants of the Prank algorithm. Moreover, it is useful to most review sites because we use only rating and review data.

We propose a system for detecting local events in the real-world using geolocation information from microblog documents. A local event happens when people with a common purpose gather at the same time and place. To detect such an event, we identify a group of Twitter documents describing the same theme that were generated within a short time and a small geographic area. Timestamps and geotags are useful for finding such documents, but only 0.7% of documents are geotagged and not sufficient for this purpose. Therefore, we propose an automatic geotagging method that identifies the location of non-geotagged documents. Our geotagging method successfully increased the number of geographic groups by about 115 times. For each group of documents, we extract co-occurring terms to identify its theme and determine whether it is about an event. We subjectively evaluated the precision of our detected local events and found that it had 25.5% accuracy. These results demonstrate that our system can detect local events that are difficult to identify using existing event detection methods. A user can interactively specify the size of a desired event by manipulating the parameters of date, area size, and the minimum number of Twitter users associated with the location. Our system allows users to enjoy the novel experience of finding a local event happening near their current location in real time.

We developed a simple method of improving the accuracy of rating prediction using feature words extracted from customer reviews. Many rating predictors work well for a small and dense dataset of customer reviews. However, a practical dataset tends to be large and sparse, because it often includes too many products for each customer to buy and evaluate. Data sparseness reduces prediction accuracy. To improve accuracy, we reduced the dimension of the feature vector using feature words extracted by analyzing the relationship between ratings and accompanying review comments instead of using ratings. We applied our method to the Pranking algorithm and evaluated it on a corpus of golf course reviews supplied by a Japanese e-commerce company. We found that by successfully reducing data sparseness, our method improves prediction accuracy as measured using RankLoss.

Abstract We present a method for synthesizing fluid animation from a single image, using a fluid video database. The user inputs a target painting or photograph of a fluid scene along with its alpha matte that extracts the fluid region of interest in the scene. Our approach allows the user to generate a fluid animation from the input image and to enter a few additional commands about fluid orientation or speed. Employing the database of fluid examples, the core algorithm in our method then automatically assigns fluid videos for each part of the target image. Our method can therefore deal with various paintings and photographs of a river, waterfall, fire, and smoke. The resulting animations demonstrate that our method is more powerful and efficient than our prior work.

Abstract We propose a system that allows the user to design a continuous flow animation starting from a still fluid image. The basic idea is to apply the fluid motion extracted from a video example to the target image. The system first decomposes the video example into three components, an average image, a flow field and residuals. The user then specifies equivalent information over the target image. The user manually paints the rough flow field, and the system automatically refines it using the estimated gradients of the target image. The user semi‐automatically transfers the residuals onto the target image. The system then approximates the average image and synthesizes an animation on the target image by adding the transferred residuals and warping them according to the user‐specified flow field. Finally, the system adjusts the appearance of the resulting animation by applying histogram matching. We designed animations of various pictures, such as rivers, waterfalls, fires, and smoke.

Effective digital content creation tools must be both efficient in the interactions they provide but also allow full user control. There may be occasions, when art direction requires changes that contradict physical laws. In particular, it is known that physical correctness of reflections for the human observer is hard to assess. For many centuries, traditional artists have exploited this fact to depict reflections that lie outside the realm of physical possibility. However, a system that gives explicit control of this effect to digital artists has not yet been described. This paper introduces a system that transforms physically correct reflections into art-directed reflections, as specified by reflection constraints . The system introduces a taxonomy of reflection editing operations, using an intuitive user interface, that works directly on the reflecting surfaces with real-time visual feedback using a GPU. A user study shows how such a system can allow users to quickly manipulate reflections according to an art direction task.

We present a method for representing solid objects with spatially-varying oriented textures by repeatedly pasting solid texture exemplars. The underlying concept is to extend the 2D texture patch-pasting approach of lapped textures to 3D solids using a tetrahedral mesh and 3D texture patches. The system places texture patches according to the user-defined volumetric tensor fields over the mesh to represent oriented textures. We have also extended the original technique to handle nonhomogeneous textures for creating solid models whose textural patterns change gradually along the depth fields. We identify several texture types considering the amount of anisotropy and spatial variation and provide a tailored user interface for each. With our simple framework, large-scale realistic solid models can be created easily with little memory and computational cost. We demonstrate the effectiveness of our approach with several examples including trees, fruits, and vegetables.

We present an appearance-based user interface for artists to efficiently design customized image-based lighting environments. 1 Our approach avoids typical iterations of parameter editing, rendering, and confirmation by providing a set of intuitive user interfaces for directly specifying the desired appearance of the model in the scene. Then the system automatically creates the lighting environment by solving the inverse shading problem. To obtain a realistic image, all-frequency lighting is used with a spherical radial basis function (SRBF) representation. Rendering is performed using precomputed radiance transfer (PRT) to achieve a responsive speed. User experiments demonstrated the effectiveness of the proposed system compared to a previous approach.

In the last decade, the vision of future interfaces has shifted from virtual reality to augmented and tangible user interfaces (UI) where virtual and physical (or "bits and atoms") co-exist in harmony. Recently, a growing number of designers and researchers have been taking the next logical step: creating interfaces where physical, tangible elements are not merely dynamically coupled to the digital attributes and information, but are themselves dynamic, self-reconfigurable devices that can change their physical properties depending on the state of the interfaces, the user, or the environment.A combination of the actuation, self-configuration, and tangibility can expand and enhance the design of tangible interfaces. In this paper, we present an overview of the use of actuation in user interfaces and discuss the rationality of building actuated interfaces. We then discuss actuated interfaces in detail based on our experience designing Lumen shape displays. Work on actuated interfaces is still in its infancy, projects are few and far between, so we consider this paper an invitation to discussion and hope it can help stimulate further research in this area.

We present a system for modeling flowers in three dimensions quickly and easily while preserving correct botanical structures. We use floral diagrams and inflorescences , which were developed by botanists to concisely describe structural information of flowers. Floral diagrams represent the layout of floral components on a single flower, while inflorescences are arrangements of multiple flowers. Based on these notions, we created a simple user interface that is specially tailored to flower editing, while retaining a maximum variety of generable models. We also provide sketching interfaces to define the geometries of floral components. Separation of structural editing and editing of geometry makes the authoring process more flexible and efficient. We found that even novice users could easily design various flower models using our technique. Our system is an example of application-customized sketching, illustrating the potential power of a sketching interface that is carefully designed for a specific application.

This paper presents an interactive system for designing and browsing volumetric illustrations. Volumetric illustrations are 3D models with internal textures that the user can browse by cutting the models at desired locations. To assign internal textures to a surface mesh, the designer cuts the mesh and provides simple guiding information to specify the correspondence between the cross-section and a reference 2D image. The guiding information is stored with the geometry and used during the synthesis of cross-sectional textures. The key idea is to synthesize a plausible cross-sectional image using a 2D texture-synthesis technique, instead of sampling from a complete 3D RGB volumetric representation directly. This simplifies the design interface and reduces the amount of data, making it possible for non-experts to rapidly design and use volumetric illustrations. We believe that our system can enrich human communications in various domains, such as medicine, biology, and geology.

We will present a user interface for quickly and easily modeling three-dimensional (3D) botanical trees from freehand sketches. The system generates 3D geometry from a two-dimensional (2D) sketch based on the assumption that trees spread their branches uniformly. Several editing operations are implemented, such as adding, cutting, and erasing branches. Our system also predicts, and generates, branches automatically. The user can continue editing a 3D tree by selecting the predicted branches. The interface of our system allows novices to design reasonably natural-looking trees quickly and interactively as compared with the interfaces of rule-based systems designed for expert users (e.g., L-systems).