Article
Computer Science, Software Engineering
Nicholas Sharp, Alec Jacobson
Summary: This paper introduces a neural implicit representation method that allows for direct geometric queries on a wide range of architectures. By applying range analysis to neural networks, the output range is bounded, and various geometric queries are developed. These queries can be efficiently evaluated on GPUs and provide accuracy guarantees, even on randomly-initialized networks.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Computer Science, Software Engineering
Kavosh Jazar, Paul G. Kry
Summary: This paper utilizes the temporal coherence of closed-form animated implicit surfaces to save resources in static or slowly-evolving areas by locally re-evaluating the implicit field. It treats implicit surface rendering as a special case of the continuous constraint satisfaction problem and proposes a temporally-coherent set inversion algorithm. By applying this algorithm on the GPU, significant speedups are achieved in complex scenes with localized deformations.
ACM TRANSACTIONS ON GRAPHICS
(2023)
Article
Computer Science, Software Engineering
Yusen Wang, Zongcheng Li, Yu Jiang, Kaixuan Zhou, Tuo Cao, Yanping Fu, Chunxia Xiao
Summary: We present a novel neural surface reconstruction method called NeuralRoom, which can reconstruct room-sized indoor scenes directly from a set of 2D images. The method reduces the shape-radiance ambiguity of implicit neural surfaces by introducing reliable geometric priors and improves the accuracy and completeness of flat regions using the perturbation-residual restrictions method.
ACM TRANSACTIONS ON GRAPHICS
(2022)
Article
Multidisciplinary Sciences
Muhammed Veli, Deniz Mengu, Nezih T. Yardimci, Yi Luo, Jingxi Li, Yair Rivenson, Mona Jarrahi, Aydogan Ozcan
Summary: Recent advances in deep learning have provided non-intuitive solutions to inverse problems in optics. Diffractive networks merge wave-optics with deep learning to design task-specific elements for all-optically performing tasks like object classification. The authors present a diffractive network used to shape broadband pulses into desired optical waveforms, demonstrating direct pulse shaping in the terahertz spectrum.
NATURE COMMUNICATIONS
(2021)
Article
Environmental Sciences
Yingjie Qu, Fei Deng
Summary: The paper proposes a one-stage method for generating 3D mesh models directly from multi-view satellite imagery. The method utilizes a continuous signed distance function (SDF) and volume rendering framework to learn the SDF values. To address lighting variations and inconsistent appearances, a latent vector and multi-view stereo constraint are incorporated. Experimental results show that the method achieves superior visual quality and quantitative accuracy in generating mesh models, and can also learn seasonal variations in satellite imagery.
Article
Polymer Science
WooSeok Choi, Sungchan Yun
Summary: Controlling the residence time of drops on solid surfaces is crucial for engineering applications such as self-cleaning and anti-icing. The relationship between drop shape and surface curvature significantly affects the bouncing characteristics, with the principle of adjusting residence time based on ellipsoidal shapes and initial surface curvature elucidated through momentum asymmetry.
Article
Biophysics
Mikhail Suyetin, Stefan Rauwolf, Sebastian Patrick Schwaminger, Chiara Turrina, Leonie Wittmann, Saientan Bag, Sonja Berensmeier, Wolfgang Wenzel
Summary: Understanding the interactions between proteins and silica surfaces is crucial for various applications, and this study uses the EISM model to investigate peptide-silica interactions. Computational screening and experimental data confirm the accuracy of the EISM model in predicting peptide-surface interactions. The results provide guidance for future experimental and theoretical research.
COLLOIDS AND SURFACES B-BIOINTERFACES
(2022)
Article
Computer Science, Artificial Intelligence
Lucas de Vries, Rudolf L. M. van Herten, Jan W. Hoving, Ivana Isgum, Bart J. Emmer, Charles B. L. M. Majoie, Henk A. Marquering, Efstratios Gavves
Summary: This paper proposes a novel approach for CT perfusion analysis using physics-informed learning, which accurately estimates perfusion parameters of affected cerebral tissue. The method is validated on simulated and real patient data, showing its effectiveness and reliability.
MEDICAL IMAGE ANALYSIS
(2023)
Article
Automation & Control Systems
Minsik Seo, Seungjae Min
Summary: This paper introduces a deep neural network-based method for accelerating topology optimization in irregular design domains, which can predict (near-)optimal topologies.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Engineering, Civil
Yiping Xie, Nils Bore, John Folkesson
Summary: This article presents a method for reconstructing high-quality bathymetry using sidescan sonar data. It leverages deep learning to estimate surface normals and uses an optimization framework for bathymetric map reconstruction. By fusing multiple observations from different sidescan lines, the estimated results are improved through optimization.
IEEE JOURNAL OF OCEANIC ENGINEERING
(2023)
Article
Computer Science, Information Systems
Tao Peng, Jianquan Lu, Zhengwen Tu, Jungang Lou
Summary: In this study, a non-decomposing method, namely the implicit Lyapunov function method, is proposed for the finite-time stabilization of quaternion-valued neural networks with time delays. The method offers relaxed constraints on time delays, flexible controllers without sign functions, and more general activation functions. The properties of the implicit Lyapunov function, such as positive definiteness, monotonicity, and radial unboundedness, are analyzed. Sufficient conditions for the finite-time stabilization of the networks are provided, and an improved adaptive control strategy is designed to enhance the stabilization efficiency. Numerical examples are presented to demonstrate the correctness, applicability, and effectiveness of the method.
INFORMATION SCIENCES
(2022)
Article
Computer Science, Information Systems
Xiaolong Jiang, Heli Sun, Yuan Chen, Liang He
Summary: Groups where like-minded people gather to share interests, comments, or participate in activities have gained popularity in social platforms. For event-based group recommendations, traditional methods fail to consider social relationships and multiple attributes of groups. This study proposes a knowledge-enhanced trust propagation graph neural network (KTPGN) that predicts user preferences by exploiting social trust, group attributes, and user-group interaction history. Experimental results on Meetup datasets show that KTPGN outperforms existing recommendation methods.
INFORMATION SCIENCES
(2023)
Article
Environmental Sciences
Mosaad Khadr, Andreas Schlenkhoff
Summary: The study introduces an optimization-simulation framework using implicit stochastic optimization, genetic algorithms, and recurrent neural networks to improve reservoir management. Results from the application at the Bigge reservoir in Germany show promising effectiveness of the GA-ISO-RNN model in simulating and predicting optimal reservoir release.
ENVIRONMENTAL SCIENCE AND POLLUTION RESEARCH
(2021)
Article
Physics, Mathematical
Ebtsam H. Taha
Summary: This article establishes the necessary and sufficient conditions for a Finsler surface to be Landsbregian through studying the Berwald curvature 2-form. The author investigates various Finsler surfaces that satisfy certain flag curvature conditions and determines when they are Riemannian. The research presented in the article has important implications on rigidity results.
INTERNATIONAL JOURNAL OF GEOMETRIC METHODS IN MODERN PHYSICS
(2023)
Article
Computer Science, Software Engineering
T. V. Christiansen, J. A. Baerentzen, R. R. Paulsen, M. R. Hannemose
Summary: Neural implicit surfaces are effective for representing shapes with arbitrary topology, but open surfaces are still challenging. The generalized winding number (GWN) is a promising approach for distinguishing points on 3D shapes. However, it lacks information about the distance to the surface, which is necessary for tasks like ray tracing. To address this, we propose the semi-signed distance field (SSDF) representation, which combines the GWN and surface distance. We compare the GWN and SSDF for various applications and find that both are suitable for neural representation of open surfaces.
COMPUTER GRAPHICS FORUM
(2023)
Article
Mathematics, Applied
Tiago Novello, Joao Paixao, Carlos Tomei, Thomas Lewiner
Summary: This paper introduces the concept of discrete line fields and derives some basic results through the study of discrete line fields, including the Euler-Poincare formula, Morse-Smale decomposition, and topologically consistent cancellation of critical elements, which allows for topological simplification of the original discrete line field.
TOPOLOGY AND ITS APPLICATIONS
(2021)
Article
Computer Science, Software Engineering
Hallison Paz, Daniel Perazzo, Tiago Novello, Guilherme Schardong, Luiz Schirmer, Vinicius da Silva, Daniel Yukimura, Fabio Chagas, Helio Lopes, Luiz Velho
Summary: We introduce MR-Net, a general architecture for multiresolution sinusoidal neural networks, and a framework for imaging applications based on this architecture. Our experiments show that MR-Net can faithfully represent the effect of sequentially applying low-pass filters in a high-resolution image.
COMPUTERS & GRAPHICS-UK
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Hallison Paz, Tiago Novello, Vinicius Silva, Guilherme Schardong, Luiz Schirmer, Fabio Chagas, Helio Lopes, Luiz Velho
Summary: MR-Net is a general architecture for multiresolution neural networks, which enables continuous representation of images in both space and scale, and it is a compact and efficient representation. By demonstrating applications such as texture magnification, minification, and antialiasing, we show the practical value of this architecture.
2022 35TH SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI 2022)
(2022)
Proceedings Paper
Computer Science, Information Systems
Pedro Torres, Andre Davys, Thuener Silva, Luiz Schirmer, Andre Kuramoto, Bruno Itagyba, Cristiane Salgado, Sidney Comandulli, Patricia Ventura, Leonardo Fialho, Marinho Fischer, Marcos Kalinowski, Simone Barbosa, Helio Lopes
Summary: Workers in large industries are exposed to various hazards, with PPE being essential for their protection. Monitoring systems using CCTV cameras for real-time detection of PPE usage are crucial. This paper proposes a novel cognitive safety analysis component based on deep learning techniques, providing robust and effective results.
PROCEEDINGS OF THE 23RD INTERNATIONAL CONFERENCE ON ENTERPRISE INFORMATION SYSTEMS (ICEIS 2021), VOL 1
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Luiz Schirmer, Guilherme Schardong, Vinicius da Silva, Helio Lopes, Tiago Novello, Daniel Yukimura, Thales Magalhaes, Hallison Paz, Luiz Velho
Summary: This survey introduces methods using neural networks for implicit representations of 3D geometry, exploring their applications in shape modeling and synthesis. It aims to provide a theoretical analysis of 3D shape reconstruction using deep neural networks and facilitate discussions among researchers in this field.
2021 34TH SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI 2021)
(2021)
Proceedings Paper
Computer Science, Artificial Intelligence
Luiz Schirmer, Djalma Lucio, Leandro Cruz, Alberto Raposo, Luiz Velho, Helio Lopes
Summary: A novel gating mechanism, Semantic Graph Attention, is proposed for 3D applications to improve the performance of Semantic Graph Convolutions by exploring channel-wise inter-dependencies. The proposed method performs the unprojection of 2D points onto their 3D version, used for estimating 3D human pose from 2D images. The attention layer improves skeleton estimation accuracy with 58% fewer parameters than the state-of-the-art.
2021 34TH SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES (SIBGRAPI 2021)
(2021)
Article
Computer Science, Software Engineering
Rafael Maio, Tiago Araujo, Bernardo Marques, Andre Santos, Pedro Ramalho, Duarte Almeida, Paulo Dias, Beatriz Sousa Santos
Summary: Augmented Reality (AR) is a crucial technology in Industry 4.0 and smart manufacturing, particularly in the field of data monitoring. In this study, we developed a Pervasive AR tool for data monitoring, along with a web application for comparison purposes. User studies were conducted to gather feedback and evaluate the effectiveness of the systems, confirming the potential of Pervasive AR for data monitoring.
COMPUTERS & GRAPHICS-UK
(2024)
Article
Computer Science, Software Engineering
Berk Cebeci, Mehmet Bahadir Askin, Tolga K. Capin, Ufuk Celikcan
Summary: Despite advances in virtual reality technologies, extended VR sessions with head-mounted displays (HMDs) still face challenges in terms of comfort. In this study, a methodology using gaze-directed and visual saliency-guided paradigms for automatic stereo camera control in real-time interactive VR was proposed. The results showed that the gaze-directed approach outperformed the saliency-guided approach, both improving the overall depth feeling without hindering visual comfort in the tested virtual environments (VEs).
COMPUTERS & GRAPHICS-UK
(2024)
Article
Computer Science, Software Engineering
Ali Egemen Tasoren, Ufuk Celikcan
Summary: By developing the NOVAction engine, we have created the NOVAction23 dataset, which consists of highly diversified and photorealistic synthetic human action sequences. This dataset is significant in improving the performance of human action recognition.
COMPUTERS & GRAPHICS-UK
(2024)
