Article
Chemistry, Analytical
Tingyu Zhang, Jian Wang, Xinyu Yang
Summary: In recent years, point cloud-based 3D object detection has achieved great success. However, previous methods did not fully consider density variation in point sampling and feature extraction. This paper proposes a new method called DSASA, which investigates point density in the sampling process and enhances point features using raw point coordinates. Experimental results on the KITTI dataset demonstrate the superiority of DSASA.
Article
Computer Science, Information Systems
Erfan Farhangi Maleki, Lena Mashayekhy, Seyed Morteza Nabavinejad
Summary: In this paper, the authors propose efficient offloading approaches for edge computing, taking into account spatio-temporal uncertainties and dynamics. They formulate the problem as an integer programming model and use machine learning to predict future locations and specifications. The proposed approaches, S-OAMC and G-OAMC, achieve near-optimal turnaround time and low migration rates.
IEEE TRANSACTIONS ON MOBILE COMPUTING
(2023)
Article
Computer Science, Artificial Intelligence
Hao Liu, Hui Yuan, Junhui Hou, Raouf Hamzaoui, Wei Gao
Summary: We propose a generative adversarial network for point cloud upsampling that evenly distributes points on the underlying surface and efficiently generates clean high-frequency regions. The network includes a dynamic graph hierarchical residual aggregation unit and a hierarchical residual aggregation unit for point feature extraction and upsampling. The discriminator uses a graph filter to extract and retain high-frequency points and generate neat edges. Experimental results show that our method generates upsampled point clouds of better quality compared to existing techniques.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Computer Science, Software Engineering
Bing Han, Lixiang Deng, Yi Zheng, Shuang Ren
Summary: This paper proposes a novel unsupervised upsampling network called UPU-SNet, which consists of two branches based on hierarchical spatial-aware transformers. The network can generate denser and more uniform point clouds through multi-level feature extraction and optimization. Extensive experiments show that the proposed approach outperforms existing unsupervised methods and achieves competitive results against previous supervised and self-supervised methods.
COMPUTERS & GRAPHICS-UK
(2023)
Article
Engineering, Electrical & Electronic
Pingping Zhang, Xu Wang, Lin Ma, Shiqi Wang, Sam Kwong, Jianmin Jiang
Summary: This paper proposes a novel progressive point cloud upsampling framework to address the non-uniform distribution issue. The model incorporates a feature expansion module and a hybrid loss function, achieving state-of-the-art performance in both qualitative and quantitative evaluations.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Engineering, Electrical & Electronic
Yoon Hak Kim
Summary: In this paper, we consider a scenario where nodes of a graph are sampled for bandlimited graph signals and uniformly quantized with optimal rate. We propose a greedy algorithm to construct the best sampling set that minimizes the average reconstruction error. By using QR factorization, we manipulate the reconstruction error and provide an analytic result that the next minimizing node can be iteratively selected by minimizing the geometric mean of the row vectors of the inverse upper triangular matrix R(-1) in the QR factorization. We compare the complexity of the proposed algorithm with different sampling methods and evaluate its performance through experiments, demonstrating its superiority in the presence of quantization.
EURASIP JOURNAL ON ADVANCES IN SIGNAL PROCESSING
(2022)
Article
Computer Science, Information Systems
Hengrun Zhang, Kai Zeng
Summary: This article proposes a communication-aware secret share placement strategy to optimize communication overhead in a hierarchical edge computing architecture while guaranteeing privacy constraints. The constructed optimization problem is shown to be NP-hard, and efficient heuristic algorithms are applied to find suboptimal solutions. Comprehensive experimental results demonstrate the advantage of the proposed algorithms.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Computer Science, Software Engineering
Martin Skrodzki, Eric Zimmermann
Summary: This paper introduces a weighting scheme for neighborhoods in point sets that considers the geometry shape and normal information, making the obtained neighborhoods more reliable and dependent on the orientation of the point set. By utilizing a sigmoid to define weights based on normal variation and evaluating them on large models, optimal parameters selection and the inapplicability of globally fixed neighborhood sizes for model processing are revealed. The applicability of the weighting scheme in the context of denoising is highlighted.
COMPUTER-AIDED DESIGN
(2021)
Article
Computer Science, Artificial Intelligence
Hogeon Seo, Sangjun Noh, Sungho Shin, Kyoobin Lee
Summary: This study proposes Probability Propagation (PP) as a stochastic upsampling method to improve the performance and efficiency of neural networks (NN) in point cloud segmentation (PCS). By replacing the iterative inference of NN with PP, large point clouds can be dealt with quickly and efficiently. Experimental results demonstrate that using NN+PP achieves higher performance and faster inference speed compared to when using NN alone, indicating the significant contribution of PP in PCS when used in edge AI systems.
PATTERN RECOGNITION LETTERS
(2023)
Article
Environmental Sciences
Weite Li, Kyoko Hasegawa, Liang Li, Akihiro Tsukamoto, Satoshi Tanaka
Summary: The proposed deep learning-based upsampling method improves the distributional density, uniformity, and connectivity of points in the edge regions of 3D-scanned point clouds. The method effectively increases the percentage of edge points and works well for sharp and soft edges, with good results when combined with a noise-eliminating filter. This approach demonstrates effectiveness in enhancing the visibility of edge-highlighted transparent visualization of complex 3D-scanned objects.
Article
Computer Science, Artificial Intelligence
Bing Han, Xinyun Zhang, Shuang Ren
Summary: This paper presents a point cloud upsampling technique that captures the global and local structured features of the point cloud using a feature extractor and feature expander. The proposed method outperforms previous methods in network performance for 3D point cloud upsampling, achieving more efficient inference with fewer parameters.
IMAGE AND VISION COMPUTING
(2022)
Article
Computer Science, Information Systems
Shangguang Wang, Yan Guo, Ning Zhang, Peng Yang, Ao Zhou, Xuemin Shen
Summary: This article investigates microservice coordination among edge clouds to enable seamless and real-time responses to service requests from mobile users. Through proposing different algorithms for optimization experiments, it is proven that the online algorithm's performance is close to the optimal performance obtained by the offline algorithm.
IEEE TRANSACTIONS ON MOBILE COMPUTING
(2021)
Article
Computer Science, Information Systems
Seungkyun Lee, SuKyoung Lee, Seung-Seob Lee
Summary: A task scheduling algorithm in the CEC environment is proposed in this study, taking into account task priority and network flows to reduce the deadline miss ratio of IoT applications. Simulation results demonstrate that the proposed algorithm can significantly reduce the DMR compared to other benchmark methods.
IEEE WIRELESS COMMUNICATIONS LETTERS
(2021)
Article
Computer Science, Information Systems
Dimitrios Spatharakis, Marios Avgeris, Nikolaos Athanasopoulos, Dimitrios Dechouniotis, Symeon Papavassiliou
Summary: The evolution of IIoT and edge computing allows resource-constrained mobile robots to offload computationally intensive localization algorithms. This poses a challenge in joint co-design of communication, control, estimation, and computing infrastructure when utilizing remote resources. A set-based estimation offloading framework is introduced for unicycle robot navigation, accounting for modeling and measurement uncertainties. A switching set-based control mechanism and utility-based offloading mechanism are designed to ensure accurate navigation and optimize resource utilization.
IEEE INTERNET OF THINGS JOURNAL
(2023)
Article
Computer Science, Information Systems
Zhiqing Tang, Jiong Lou, Weijia Jia
Summary: Due to its lightweight and easy deployment, the use of containers has become a promising approach for Mobile Edge Computing (MEC). However, existing work has neglected the fact that scheduling tasks at the level of layers instead of images can greatly reduce task completion time in resource-limited MEC. To address this gap, a novel layer dependency-aware container scheduling algorithm is proposed, taking into account the complex dependency between layers and images.
IEEE TRANSACTIONS ON MOBILE COMPUTING
(2023)
Article
Computer Science, Software Engineering
Wendong Mao, Mingjie Wang, Hui Huang, Minglun Gong
Summary: Two neural networks are proposed in this study: a matching network trained on binocular stereo datasets to handle textureless regions and a second network that consolidates 3D point clouds through point projection for aligning individual patches. Together, these networks show comparable results to existing state-of-the-art approaches on the DTU dataset.
Article
Computer Science, Information Systems
Mingjie Wang, Hao Cai, Xian-Feng Han, Jun Zhou, Minglun Gong
Summary: This paper proposes a novel network called STNet for accurate crowd counting. STNet consists of two key components: Scale-Tree Diversity Enhancer and Multi-level Auxiliator. By enriching scale diversity and exploiting shared characteristics at multiple levels, STNet can significantly improve the accuracy of crowd counting.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Software Engineering
Min Lu, Joel Lanir, Chufeng Wang, Yucong Yao, Wen Zhang, Oliver Deussen, Hui Huang
Summary: The article focuses on the use of perceptual laws to model and perceive small differences between visual elements, particularly exploring the effects of distance and intensity as major visual variables in this perception.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2022)
Article
Computer Science, Software Engineering
Xinxin Zhang, Yuefeng Xi, Zhentao Huang, Lintao Zheng, Hui Huang, Yueshan Xiong, Kai Xu
Summary: This paper proposes a novel high-accuracy active hand-eye calibration approach that improves the calibration accuracy through guiding robot movement and camera view selection. The method employs an online estimated discrete viewing quality field to guide data acquisition and selects the next-best-view based on view quality. Experimental results show that the algorithm outperforms other approaches in terms of accuracy and robustness.
Article
Engineering, Electrical & Electronic
Qingquan Li, Hui Huang, Wenshuai Yu, San Jiang
Summary: Unmanned aerial vehicles have become widely used in remote sensing and are critical in the construction of smart cities. However, urban environments pose challenges for secure and accurate data acquisition for 3D modeling. This study presents optimized views photogrammetry as a solution and verifies its precision and potential in large-scale 3D modeling.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Computer Science, Information Systems
Liqiang Lin, Pengdi Huang, Chi-Wing Fu, Kai Xu, Hao Zhang, Hui Huang
Summary: We propose a new attention-based mechanism for learning enhanced point features in point cloud processing tasks. Unlike previous studies, our approach learns to locate the best attention points to optimize the performance of specific tasks. We advocate the use of single attention points for better semantic understanding in point feature learning.
SCIENCE CHINA-INFORMATION SCIENCES
(2023)
Article
Agricultural Engineering
Jiawei Li, Weihong Ma, Qiang Bai, Dan Tulpan, Minglun Gong, Yi Sun, Xianglong Xue, Chunjiang Zhao, Qifeng Li
Summary: This study proposes an automatic method for measuring the body size parameters of beef cattle, which reduces the influence of postures on measurement results through key region segmentation, body size calculation, and data calibration. Experimental results show that the adjustment model significantly minimizes measurement errors and improves the accuracy of body size measurement.
BIOSYSTEMS ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Xin Huang, Minglun Gong
Summary: Face aging is an active research field in multimedia applications, but current approaches still struggle with generating convincing age progression while preserving personal identity. This paper proposes a novel approach called Landmark-guided Dual-learning cGAN (LDcGAN) with a multi-attention mechanism to address these limitations. The LDcGAN uses external landmark attention to adjust facial structure variations and built-in attention to emphasize discriminative regions relevant to aging. It achieves improved age consistency and minimal changes to personal identity and background, producing appealing results in terms of image quality, personal identity, and age accuracy.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Software Engineering
Yilin Liu, Ruiqi Cui, Ke Xie, Minglun Gong, Hui Huang
Summary: Traditional urban reconstruction methods can only output incomplete 3D models, while learning-based shape reconstruction techniques are designed for single objects. This paper proposes a novel learning-based approach for real-time complete 3D mesh reconstruction of large-scale urban scenes. The approach segments objects and determines their positions, and reconstructs them under local coordinates to approximate training datasets.
COMPUTERS & GRAPHICS-UK
(2023)
Article
Computer Science, Artificial Intelligence
Mingjie Wang, Jun Zhou, Hao Cai, Minglun Gong
Summary: This paper proposes a novel and efficient counting model, CrowdMLP, which regresses total counts by designing a multi-granularity MLP regressor that models global dependencies of embeddings. The model uses a locally-focused pre-trained frontend to extract crude feature maps with spatial cues and tokenizes the crude embeddings and raw crowd scenes at different granularities. The study also introduces a self-supervised proxy task, Split-Counting, to overcome limited samples and the lack of spatial hints.
PATTERN RECOGNITION
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Mingjie Wang, Hao Cai, Yong Dai, Minglun Gong
Summary: Crowd counting has become a topic of increasing interest due to its challenges and wide applications. Existing methods often require labor-intensive location-level annotations, hindering the generalization of location-adherent models. To address this issue, this paper proposes a novel Dynamic Mixture of Counter Network (DMCNet) for location-agnostic crowd counting.
2023 IEEE/CVF WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV)
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Zimu Yi, Ke Xie, Jiahui Lyu, Minglun Gong, Hui Huang
Summary: The use of image-based rendering (IBR) technique is important for implementing VR telepresence by allowing interactive presentation of real scenes to viewers. However, the quality of IBR results depends on various factors such as pre-captured views and rendering algorithms. In this work, we introduce the concept of renderability, which predicts the quality of IBR results at any given viewpoint and view direction, to guide the selection of viewpoints/trajectories for challenging large-scale 3D scenes.
2023 IEEE CONFERENCE VIRTUAL REALITY AND 3D USER INTERFACES, VR
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Xingguang Yan, Liqiang Lin, Niloy J. Mitra, Dani Lischinski, Daniel Cohen-Or, Hui Huang
Summary: ShapeFormer is a transformer-based network that produces a distribution of object completions from incomplete and noisy point clouds. It introduces a compact 3D representation called VQDIF to facilitate the use of transformers in 3D. Experimental results show that ShapeFormer outperforms previous methods in terms of completion quality and diversity, and it effectively handles various shape types and real-world scans.
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022)
(2022)