WSCG '2018: Short Papers Proceedings

Permanent URI for this collection

http://hdl.handle.net/11025/34642

Browse

Showing 1 - 20 out of 25 results

Empirical study on label smoothing in neural networks
(Václav Skala - UNION Agency, 2018) Mezzini, Mauro; Skala, Václav
Neural networks are now day routinely employed in the classification of sets of objects, which consists in predicting the class label of an object. The softmax function is a popular choice of the output function in neural networks. It is a probability distribution of the class labels and the label with maximum probability represents the prediction of the neural network, given the object being classified. The softmax function is also used to compute the loss function, which evaluates the error made by the network in the classification task. In this paper we consider a simple modification to the loss function, called label smoothing. We experimented this modification by training a neural network using 12 data sets, all containing a total of about 1:5 106 images. We show that this modification allow a neural network to achieve a better accuracy in the classification task.
AgeRegression: rejuvenating 3D-facial scans
(Václav Skala - UNION Agency, 2018) Legde, Katharina; Castillo, Susana; Cunningham, Douglas W.; Skala, Václav
The majority of virtual agents have adult bodies. There are, however, a number of reasons for using younger avatars. For example, an adult interface agent usually leads users to expect adult-level communicational and social skills. As a result, users tend to be rather intolerant when the interface agent makes obvious mistakes (e.g., incorrect grammar) or uses inappropriate behavior (e.g., looking away from the interlocutor). Since computer social-skills are still under-developed, it seems reasonable to use a body model that reflects this: child avatars. Unfortunately, the use of database-driven techniques for creating a variable-aged animation system would require a very large number of scans of children at different ages, making such a system impractical for technical and ethical reasons. As an alternative, this paper develops and validates a method for synthetically and systematically altering the apparent age of a virtual character. The here proposed technique is able to create younger and older versions of a facial scan and guarantees that the resulting meshes can be animated. Starting with a three-dimensional, adult facial scan, we use a physiologically-inspired, trigonometric polynomial to age-regress the model to a desired age. Quantitative measurements show that the technique can reconstruct the correct anthropometric proportions of 2-10 year-old children. A perceptual experiment provides an initial mapping of the technique’s parameters onto the perceived age and realism.
Structural identification of crystal lattices based on fuzzy neural network approach
(Václav Skala - UNION Agency, 2018) Kirsh, Dmitriy; Kupriyanov, Alexandr; Paringer, Rustam; Skala, Václav
Each crystal nanostructure consists of a set of minimal building blocks (unit cells) which parameters comprehensively describe the location of atoms or atom groups in a crystal. However, structure recognition is greatly complicated by the ambiguity of unit cell choice. To solve the problem, we propose a new approach to structural identification of crystal lattices based on fuzzy neural networks. The paper deals with the Takagi- Sugeno-Kang model of fuzzy neural networks. Moreover, a three-stage neural network learning process is presented: in the first two stages crystal lattices are grouped in non-overlapping classes, and lattices belonging to overlapping classes are recognized at the third stage. The proposed approach to structural identification of crystal lattices has shown promising results in delimiting adjacent lattice types. The structure identification failure rates decreased to 10 % on average.
Detection of change in video based on local pattern and photometric features
(Václav Skala - UNION Agency, 2018) Chan, K L; Skala, Václav
The segmentation of moving objects in video can be formulated as a background subtraction problem – the detection of change in each image frame. The background scene is learned and modeled. A pixelwise process is employed to determine whether the current pixel is similar or not to the background model. The detection of change in video is challenging due to the non-stationary background such as illumination change, background motions, etc. We propose new features for background modeling. Perception-based local ternary patterns are generated from the same color channels as well as from different color channels. Features computed from the local patterns are stored in the background model as samples. If the current pixel is classified as background, the background model is updated. Finally, we propose a probabilistic refinement to improve each change region by taking into account the spatially consistency of image features. We compare our method with various background subtraction algorithms on some video datasets. Our method can achieve 13% better performance than other methods.
RDspeed: development framework for speed-based adaptation of web content on public displays
(Václav Skala - UNION Agency, 2018) Tafreshi, Amir E. Sarabadani; Wicki, Adrian; Tröster, Gerhard; Skala, Václav
Viewers of public displays perceive the content of a display at different walking speeds. While responsive design (RD) adapts web content to different viewing contexts, so far only the characteristics of the device and recently the proximity of the viewers are taken into account. Yet, little attention has been paid to speed-based adaption of content and its potential in case of public displays. We therefore decided to develop a framework that would support speed-based adaptation of public display applications. In this paper, we present a framework called RDSpeed that allows developers and designers alike to easily utilize our speed-based adaptation technique and integrate them into their own applications. RDSpeed extends the standard RD definition by adding new media queries for each adaptation technique. Media queries have long been established as the go-to technique for developing responsive web applications when dealing with a variety of different devices. A user study was conducted to investigate the potential of our content adaptation technique, and possible use and extensions in the future. We show several example adaptive applications of RDSpeed, as well as discussing advantages and limitations of our framework as revealed by our user study.
Advances for 3D printing: remote control system and multi-material solutions
(Václav Skala - UNION Agency, 2018) Luque, Adrián; Jurado, Juan Manuel; Cárdenas, José Luis; Feito, Francisco R.; Skala, Václav
Three-dimensional printing is an emerging manufacturing technology for many applications such as rapid prototyping, biomedical engineering and industrial designs. This paper aims to provide the implantation of a low-cost and extensible system in order to real-time monitor and modify 3D printing parameters. It is formed by a webbased platform and hardware components to manage and capture the printing process. In addition, we raise an overview about material properties in order to generate multi-material prototypes of bone tissue.
Two-phase MRI brain tumor segmentation using random forests and level set methods
(Václav Skala - UNION Agency, 2018) Lefkovits, László; Lefkovits, Szidónia; Skala, Václav
Magnetic resonance images (MRI) in various modalities contain valuable information usable in medical diagnosis. Accurate delimitation of the brain tumor and its internal tissue structures is very important for the evaluation of disease progression, for studying the effects of a chosen treatment strategy and for surgical planning as well. At the same time early detection of brain tumors and the determination of their nature have long been desirable in preventive medicine. The goal of this study is to develop an intelligent software tool for quick detection and accurate segmentation of brain tumors from MR images. In this paper we describe the developed two-staged image segmentation framework. The first stage is a voxelwise classifier based on random forest (RF) algorithm. The second acquires the accurate boundaries by evolving active contours based on the level set method (LSM). The intelligent combination of two powerful segmentation algorithms ensures performances that cannot be achieved by either of these methods alone. In our work we used the MRI database created for the BraTS ’14-‘16 challenges, considered a gold standard in brain tumor segmentation. The segmentation results are compared with the winning state of the art methods presented at the Brain Tumor Segmentation Grand Challenge and Workshop (BratsTS).
Educational virtual environment based on oculus rift and leap motion devices
(Václav Skala - UNION Agency, 2018) Zilak, Matea; Car, Zeljka; Jezic, Gordan; Skala, Václav
Virtual reality (VR) technology offers numerous benefits in different application areas, especially in education. VR brings new approaches to learning that can make the education process more attractive, while at the same time learners can develop creativity and innovativeness. Despite the possible benefits that VR can offer, the use of VR is still not widespread for educational purposes. Furthermore, the potential of VR for assistive technologies in the Augmentative and Alternative Communication (AAC) domain is recognized but has not been fully exploited. In this paper, development of an elementary mathematical virtual classroom prototype based on Oculus Rift and Leap Motion devices is described. The learning concept used in the prototype was taken from the state-of-the-art AAC application for mobile devices that introduces children with the concept of quantity which is one of the preconditions for adopting the concept of number. To analyze user’s satisfaction with the application and acceptance of a new technology in general, user evaluation of the developed prototype was conducted. In general, a positive feedback from users without disabilities suggests that it makes sense to combine VR elements with education as well as AAC technologies. Contributing factors, such as the level of immersion in VR environments, unnatural behavior of virtual hands, and the level of familiarity with the VR technology, are identified as some of the most important aspects that need to be considered in the follow-up studies concerning users with disabilities (i.e. children with complex communication needs).
Perception of basic emotion blends from facial expressions of virtual characters: pure, mixed, or complex?
(Václav Skala - UNION Agency, 2018) Mäkäräinen, Meeri; Kätsyri, Jari; Takala, Tapio; Skala, Václav
As animated virtual characters in games, movies and other applications become more humanlike, it becomes more and more important to be able to imitate the complicated facial behaviour of a real human. So far, facial expression animation and research have been dominated by the basic emotions view, limited to the six universal expressions: anger, disgust, fear, joy, sadness and surprise. More complex facial expressions can be created by blending these basic emotions, but it is not clear how these blends are perceived. Are they still perceived as basic emotions or combinations of basic emotions, or are they perceived as expressions of more complex emotions? We used a series of online questionnaires to study the perception of all pairwise blends of basic emotions. The blends were produced as a sum of facial muscle activations in the two basic emotions, using a physically-based, animated face model. Our main finding is that several basic emotion blends with an opposite valence are perceived as complex emotions that are neither pure emotions nor their blends. Blends of basic emotions with a similar valence are typically perceived as pure basic emotions (e.g., a blend of anger and disgust is perceived as pure anger). Only one of the blends (joy+surprise) was perceived as a blend of two different basic emotions.
Fish motion capture with refraction synthesis
(Václav Skala - UNION Agency, 2018) Müller, Klaus; Hütwohl, Jan-Marco; Kuhnert, Klaus-Dieter; Skala, Václav
3D fish animations become more and more popular in fish behavioral research. It empowers the experimenter to design fish stimuli and their specific behavior to the experiment’s needs. The fish animation can be done manually or derived from video footage. Especially automatic fish model parameter recovery for 3D animations is not well studied yet. Here we present a novel, flexible method for this purpose. It can be used to recover position, pose, bone rotation and size from single or multiple view and for single or multiple fish. Additionally we implement a novel method to compensate the fish tank’s refraction effect and show that this method can decrease the error up to 80 %. We successfully applied the proposed method to two different data sets and recovered fish parameters out of single- and double-view video stream. A video attached to this paper demonstrates the results.
Finding similar movies: dataset, tools, and methods
(Václav Skala - UNION Agency, 2018) Leng, Hongkun; De La Cruz Paulino, Caleb; Haider, Momina; Lu, Rui; Zhou, Zhehui; Mengshoel, Ole; Brodin, Per-Erik; Forgeat, Julien; Jude, Alvin; Skala, Václav
Recommender systems are becoming ubiquitous in online commerce as well as in video-on-demand (VOD) and music streaming services. A popular form of giving recommendations is to base them on a currently selected product (or items), and provide “More Like This,” “Items Similar to This,” or “People Who Bought This also Bought” functionality. These recommendations are based on similarity computations, also known as item-item similarity computations. Such computations are typically implemented by heuristic algorithms, which may not match the perceived item-item similarity of users. In contrast, we study in this paper a data-driven approach to similarity for movies using labels crowdsourced from a previous work. Specifically, we develop four similarity methods and investigate how user-contributed labels can be used to improve similarity computations to better match user perceptions in movie recommendations. These four methods were tested against the best known method with a user experiment (n = 114) using the MovieLens 20M dataset. Our experiment showed that all our supervised methods beat the unsupervised benchmark and the differences were both statistically and practically significant. This paper’s main contributions include user evaluation of similarity methods for movies, user-contributed labels indicating movie similarities, and code for the annotation tool which can be found at http://MovieSim.org.
Combination of temporal neural networks for improved hand gesture classification
(Václav Skala - UNION Agency, 2018) Tewari, Aditya; Taetz, Bertram; Grandidier, Frédéric; Skala, Václav
Low latency detection of human-machine interactions is an important problem. This work proposes faster detection of gestures using a combination of temporal features learnt on block time input and those learnt by contextual information. The results are reported on a standard in-car hand gesture classification challenge dataset. The recurrent neural networks which learn sequential contexts are combined with 3D convolutional neural networks (C3D). We have demonstrated that a design similar to various multi-column networks, which have been successful for image classification and understanding can also improve classification performance on varying length time series. Therefore, a combination of C3D and Long-Short-Term Memory (LSTM) is utilized for classification of hand gestures. On the task of early hand gesture classification, the proposed model outperforms the the C3D model which reports best results on full gestures. It is second best and only slightly less accurate than the best performing method, on the full gesture length.
Service-based processing and provisioning of image-abstraction techniques
(Václav Skala - UNION Agency, 2018) Richter, Marvin; Söchting, Maximilian; Semmo, Amir; Döllner, J.; Trapp, M.; Skala, Václav
Digital images and image streams represent two major categories of media captured, delivered, and shared on the Web. Techniques for their analysis, classification, and processing are fundamental building blocks in today’s digital media applications ranging from mobile image transformation apps to professional digital production suites. To efficiently process such digital media (1) independent of hardware requirements, (2) at different data complexity scales, while (3) yielding high-quality results, poses several challenges for software frameworks and hardware systems, in particular for mobile devices. With respect to these aspects, using service-based architectures is a common approach to strive for. However, unlike geodata, there is currently no standard approach for service definition, implementation, and orchestration in the domain of digital images and videos. This paper presents an approach for service-based image-processing and provisioning of processing techniques by the example of imageabstraction techniques. The generality and feasibility of the proposed system is demonstrated by different client applications that have been implemented for the Android operating system, for Google’s G-Suite Software-as-a- Service infrastructure, as well as for desktop systems. The performance of the system is discussed at the example of complex, resource-intensive image-abstraction techniques, such as watercolor rendering.
Optimizing spectral fresnel reflectance for displays
(Václav Skala - UNION Agency, 2018) Metzgar, Jonathan Brian; Semwal, Sudhanshu Kumar; Skala, Václav
Approximate equations for rendering Fresnel reflectance abound in computer graphics. We take a fresh approach and consider not only the approximation but the display device as well. The sRGB color standard is finally giving way to wide color spaces such as Adobe RGB and DCI P3 which display more color. We present a preprocessing method to use measured spectral index of refraction data and the color space specification to synthesize an RGB complex index of refraction. Metals, in particular, generally require a spectral renderer, but we created a way to sample the complex index of refraction and absorptive index that acts like the color filter built into the display. Our novel contribution uses a normal distribution centered around the ideal display red, green, and blue wavelengths derived from the CIE xy coordinates and respective white point to window sample the complex index of refraction. We created a WebGL experimental platform that uses the Schlick inspired Lazanyi and Szirmay-Kalos (LSK) multispectral Fresnel approximation coupled with modern physically based BRDFs to simulate the appearance of metal. Our application can compare five different Fresnel implementations coupled with physically based Blinn- Phong and GGX microfacet models. We demonstrate that with reasonable filter widths, we eliminate the need for a spectral renderer for real-time rendering. Additionally, we utilize publicly available measurement data to simulate a variety of metals ranging from silver, gold, and copper to silicon, lead, and zinc.
A new robust and blind image watermarking scheme in frequency domain based on optimal blocks selection
(Václav Skala - UNION Agency, 2018) Tarhouni, Nesroni; Charfeddine, Maha; Ben Amar, Chokri; Skala, Václav
Image, audio and video are the first media affected by hacking due to the availability of the internet and to the high speed connection. One of the solutions to solve such problems is watermarking. Digital watermarking is the process of embedding an imperceptible and a robust signature into a digital signal. In this paper, we focus on image watermarking. We have embedded the watermark in the frequency domain using Discrete Cosine Transform. The choice of the blocks where we insert the watermark bits depends on a preprocessing study on the original and compressed-decompressed image. Then we have implemented a blind detection algorithm. We tried to enhance the security of our technique by applying an Arnold transform to the embedded watermark. Finally, we have tested the robustness of our method by applying many attacks to the watermarked images using Stirmark 3.1. The results demonstrate that our method yields a high level of imperceptibility and robustness against JPEG compression, unique and double Stirmark attacks.
Performance evaluation of face alignment algorithms on "in-the-wild" selfies
(Václav Skala - UNION Agency, 2018) Babanin, Ivan; Mashrabov, Aleksandr; Skala, Václav
Recently mobile apps, which beautify human face or apply cute masks to a human face, become very popular and gain lots of attention in media. These tasks require very precise landmarks localization to avoid "uncanny valley" effect. We introduce the new dataset of selfies, that were taken on mobile devices, and robustly evaluate and compare different state-of-the-art approaches to the task of face alignment. Evidently, our dataset allows to reliably rank face alignment algorithms that is superior to the most popular dataset in that area of research.
3D reconstruction of coronary arteries from rotational X-ray angiography
(Václav Skala - UNION Agency, 2018) Oueslati, Chaima; Mabrouk, Sabra; Ghorbel, Faouzi; Hedi Bedoui, Mohamed; Skala, Václav
X-ray angiography has been an effective modality for diagnosing coronary artery disease representing one of the leading causes of death. A 3D reconstruction of the coronary arteries is a very important step to facilitate the interpretation of angiograms. In this paper, we propose a 3D reconstruction method of the coronary arteries. In order to improve the pairwise matching performance of our approach, we introduce artificial interest points (nodes) to respect the topology variation between 3D vascular trees. New similarity measures are proposed to take into consideration the non-rigid coronary artery movement. To measure the performance of the proposed method, we evaluate the proposed method on clinical data. Results show that accuracy of vessel centerlines has the average projection error equal to 0.5 mm for 28 different patients.
Visually realistic graphical simulation of underwater cable
(Václav Skala - UNION Agency, 2018) Ganoni, Ori; Mukundan, Ramakrishnan; Green, Richard; Skala, Václav
This paper presents different modeling considerations that are important in simulating visually realistic behavior of underwater cables attached to remotely operated vehicles. The proposed methodology has been tested on highly complex models of aquatic environments created using Unreal Engine 4. Current methods and implementations of cable simulations that are widely used in computer graphics are generally suited only to light density mediums such as air. In this paper, we present modifications to the above model required for simulating neutrally buoyant cables in underwater environments. The simulation results presented in this paper successfully demonstrate different behavioral aspects of flexible variable length underwater cables and their variations with respect to modeling parameters using our proposed method.
Actor 3D reconstruction by a scene-based, visual hull guided, multi-stereovision framework
(Václav Skala - UNION Agency, 2018) Ismael, Muhannad; Ramirez Orozco, Raissel; Loscos, Céline; Prevost, Stéphanie; Remion, Yannick; Skala, Václav
This paper proposes a novel framework to produce 3D, high-precision models of humans from multi-view capture. This method’s inputs are a visual hull and several sets of multi-baseline views. For each such view set, a surface is reconstructed with a multi-baseline stereovision method, then used to carve the visual hull. Carved visual hulls from different view sets are then fused pairwise to deliver the intended 3D model. The contributions of this paper are threefold: (i) the addition of visual hull guidance to a multi-baseline stereovision method, (ii) a carving solution to a visual hull from an interpolated and smooth stereovision surface, and (iii) a fusion solution to merge differently carved volumes differing in several areas. The paper shows that the proposed approach helps recovering a high quality carved volume, a 3D representation of the human to be modelled, that is precise even for small details and in concave areas subjected to occlusion.
3D underground reconstruction for real-time and collaborative virtual reality environment
(Václav Skala - UNION Agency, 2018) Jurado, Juan Manuel; Ortega Alvarado, Lidia; Feito, Francisco R.; Skala, Václav
Urban planning has become a relevant issue to achieve the sustainable development of the modern cities. It requires ubiquitous systems including 3D GIS capabilities and high performance without many hardware requirements. This paper deals with the development of a web application to visualize, interact and explore, through 3D perspective, the underground infrastructures of our college campus. Its 3D reconstruction raises many challenges which remains to be surmounted. The main aim of this research is the 3D modeling of underground infrastructure in order to real-time analytics of its current features. In addition, this system supports virtual reality technology to interact with the GIS data into a collaborative environment available to be performed with mobile platforms.

Browse

Recent Submissions