JSPS - Tokyo Institute of Technology

A time-space trade-off black-box data structure for stacks 🔗

[Thu. May 10,  09:50] - The need to run algorithms on limited-memory devices motivated our consideration for data structure in the settings where there is only a limited amount of memory available. We provide a black-box data structure for stack algorithms, transparent to the user, with a high degree of compression at the cost of a small increase in the running time. All of our algorithms are available in C ++ (and Julia) under MIT licenses at: https://github.com/Azzaare/CompressedStacks.cpp.git

Keio University

LMML: An Environment for Forensic Data Visualization 🔗

[Wed. May 09,  10:35] - Through the autopsy of a body, we can answer a whole range of questions as to how death happened and come up with explanations and counter-measures.The presented framework, based on a mark-up language (that we dubbed “LMML”, for Legal Medicine Mark-up Language) to store, describe and arrange forensic data, aims at overcoming those issues. Thus, through the creation of an ontology for computational forensics, this framework allows quicker, smoother input of forensic data, for better automation and visualization thereof, so that it can be used by medical examiners, investigators, as well as all the people involved in the judicial courts.

Osaka University, Institute for Datability Science

Pharmaceutical Brain Image Processing and Modeling 🔗

[Thu. May 10,  11:50] - This presentation introduces one of IDS’s projects where we were tasked to discover deformations of a brain due to pharmaceutical interventions given the unique data collection method by our client’s camera machine. This data collection method presents us with many image processing based challenges such as speckled noise, flickering, guided artifacts, etc. Therefore in this talk, we will first introduce the given data, followed by some of the methods we used to process the images, and the techniques we employed to model and deform the processed brain data for fitting purposes. Finally, we all also discuss the current results, challenges, and future plans of this project.

Osaka University, Institute for Datability Science

Sharpening Detection in Image Forensics 🔗

[Wed. May 09,  10:15] - A novel method for sharpening detection will be introduced in this talk. This method employs the efficient combination of feature extraction and machine learning. The algorithm for feature extraction is based on texture analysis, and specific designed for sharpening detection. This method is validated to lead the best performance for sharpening detection for now.

Keio University

Visualization as a means of intelligence amplification 🔗

[Wed. May 09,  14:00] - At IEEE VIS 1993, F. P. Brooks, Jr. presented an inequality “IA > AI.” Visualization can be regarded as a representative means of Intelligence Amplification. In this talk, I strive to justify this inequality with a few results of our latest visualization R&D projects.

The University of Tokyo & JFLI

Recent Advances in Gradient-Domain Rendering 🔗

[Wed. May 09,  14:40] - The most common solutions to the light transport problem rely on Monte Carlo integration in the primal-domain. This talk will introduce recent advances in gradient-domain light transport to improve the convergence rate of the estimation.

The University of Tokyo & National Institute of Informatics

Region Based Image Retrieval Revisited 🔗

[Wed. May 09,  15:00] - We revisit region based image retrieval by incorporating semantic specification of objects and intuitive specification of spatial relationships. We support multiple aspects of semantic object specification with a multitask CNN  to jointly handle multi-aspect object specification. We help users specify spatial relationships among objects in an intuitive way, with recommendations by mining the search results. We can recommend spatial relationships by category names based on language prior. Moreover, object-level inverted indexing and re-ranking are instantaneous.

Osaka University, Institute for Datability Science

Better Quantization: Efficiency Through Robustness 🔗

[Wed. May 09,  15:40] - The properties of distortion-based clustering methods are well-understood when the data is well-behaved in an approximately Gaussian sense, but all statistical guarantees fall apart when heavy-tailed observations are admitted. Distributionally robust methods have been proposed, but their implementation is non-trivial and does not scale in the data dimension. We consider a new approach, which is easily implemented and has much better dependence on the dimension.

Nagoya University

Assessment of the Attractiveness of Food Photos 🔗

[Wed. May 09,  16:40] - This talk will introduce a method for estimating the attractiveness of food photos from image features to support the generation, filtering, and re-ranking of the contents.

Kyoto University

Visual Causality Exploration and its Applications 🔗

[Thu. May 10,  11:10] - In this talk, we would like to introduce our activities on visual causality exploration. They include an interactive specification of a latent variable which explains several observable variables by using a causality graph in a phenotypic expression network, and an interactive exploration of a causality between two time-varying variables defined on dicrete points.

National Institute of Informatics

Sketch-based Manga Retrieval 🔗

[Wed. May 09,  16:00] - We introduce a sketch-based retrieval method over manga images, with a large-scale manga dataset Manga109.

Keio University

Volumetric Reconstruction of Dynamic Scenes Without Storing Volumes 🔗

[Wed. May 09,  12:15] - We present a novel real-time 3D reconstruction method for dynamic scenes using multiple RGB-D cameras. Our approach uses GPU based volumetric reconstruction, but the volume is not stored in GPU main memory. Lowered memory requirements allow reconstruction of very large scenes.

Osaka University, Institute for Datability Science

Presentation of the Osaka University Institute for Datability Science

Osaka University, Institute for Datability Science

Session chair

Kyoto University

Visual Analytics System of Changing Nonlinear Interactions in Real Systems 🔗

[Thu. May 10,  10:30] - In real systems, the relationships among time series variables are not generally static; rather they are dynamically changing in a manner that is nonlinear or state-dependent. In this talk, I will talk about a visual analytic system to support the identification and interpretation of the system state from a dynamically changing graph.

University of Paris East Marne La Vallée / JFLI / Keio University

Digital Image Forensics and CG images 🔗

[Wed. May 09,  09:55] - The process to detect if an image has been falsified or is an original belongs to the field of digital image forensics. In this framework, this talk presents some tools to automatically detect if an image is a computer graphic image or a natural image. Like in some previous methods, this approach focuses on the image noise properties, and like more and more forensics technics, it is based on a deep learning framework.

​

Geometric Algebra and Quadrics

[Thu. May 10,  11:30] - Geometric algebra is a set of tools to create, transform and intersect geometric objects. This talk presents a short introduction to geometric algebra, then presents some applications to easily compute quadric surfaces (ellipsoid, paraboloids, hyperboloid, cones, spheres, planes, ....) and quadric curves (conic sections, lines, ...).

Osaka University, Institute for Datability Science & JFLI

Fifty thousand Buddha Faces: a Big Data take on an Art History project 🔗

[Thu. May 10,  12:10] - This talk introduces a new joint project started with the department of litterature, on the large scale analysis of Buddha statue pictures. It all started with one hard drive and fifty thousand pictures.

Keio University

Session chair

Sciencescope, Shibaura Institute of Technology

Introduction to Sciencescope 🔗

[Thu. May 10,  09:30] - This talk introduced by the president of Sciencescope will present the association, its activities and the coming event RCFK2018  

On Future Internet Technology: ICN, IoT and Cybersecurity 🔗

[Thu. May 10,  09:40] - Information-Centric Networking is a novel approach for Future Internet architecture. This talks aims at introducing this concept and discusses its impact on content delivery (e.g., video streaming), cybersecurity and IoT.

Kyushu University

Transparent Object Classification using Various CNN 🔗

[Wed. May 09,  11:35] - Image processing techniques recently improve a lot, whether by hand-made features, or neural networks. However, no efficient and easy to use techniques have been found yet for transparent object classification (TOC). In this talk, we propose multiple approaches to tackle this challenge using various CNN architectures. Comparing those approaches’s pros and cons, reagrding its resource consumtion and final result, can help us identify which method is the most efficient, and why.

National Institute of Informatics

Session chair

Tokyo Institute of Technology

Interactive high-dimensional visualization 🔗

[Wed. May 09,  14:20] - Interactive high-dimensional visualization technique renders complex information on a high-dimensional Euclidean space.  The user watches at the two-dimensional projection of the high-dimensional rendering and controls his/her view by changing the projection via simple user interaction.  This talk presents applications of this technique in visual analytics of multidimensional data and social networks.

Osaka University, Institute for Datability Science

Session chair

Kyushu University

VMPFusion: Variational Message Passing for dynamic 3D face reconstruction 🔗

[Wed. May 09,  11:55] - In this talk I will describe a probabilistic approach for dynamic 3D face modeling using a consumer-grade RGB-D camera. In this research my goal is to formulate a strategy to fuse noisy 3D measurements captured with a Kinect camera into a 3D facial model without relying on explicit point correspondences. We propose to tackle this challenging problem with the Variational Message Passing (VMP) algorithm, which optimize a variational distribution using a message passing procedure on a graphical model. We show the validity of our formulation with real-data experiments.

Osaka University, Cybermedia Center

Computer Vision toward Photometric-consistent Augmented Reality 🔗

[Wed. May 09,  16:20] - The photometric consistency between a real environment and virtual objects is an important issue in improving an Augmented Reality (AR) environment. To render photometrically consistent virtual objects, properties of the light sources and the floor surfaces should be estimated in real-time. Computer Vision (CV) techniques for the photometric-consistent AR are introduced in the presentation.

RIKEN

Stream graphs and links streams for the modelling of interactions over time 🔗

[Thu. May 10,  10:10] - We present a generalization of graph concepts in order to model streams of interactions over time, i.e. face-to-face contacts, email exchanges, phone calls, etc. From elementary concepts like density, clusters, or paths, we obtain more advanced concepts like cliques, degrees, clustering coefficients, or connected components. The consistency of the language we develop makes it easy to generalize higher-level objects such as quotient graphs, line graphs, k-cores, and centralities. We will present the core of this formalism, as well as current exciting research directions.

Tokyo Institute of Technology

Interactive high-dimensional visualization 🔗

[Wed. May 09,  14:20] - Interactive high-dimensional visualization technique renders complex information on a high-dimensional Euclidean space.  The user watches at the two-dimensional projection of the high-dimensional rendering and controls his/her view by changing the projection via simple user interaction.  This talk presents applications of this technique in visual analytics of multidimensional data and social networks.

Osaka University, Institute for Datability Science

Transparent Object Segmentation in a Light-field Image 🔗

[Wed. May 09,  11:15] -  In this talk, I will introduce the recent progress of transparent object segmentation in the light-field image captured by both camera array and lenslet camera.