Article
Computer Science, Artificial Intelligence
Xinya Wang, Jiayi Ma, Junjun Jiang, Xiao-Ping Zhang
Summary: This paper proposes a method for hyperspectral image super-resolution through abundance domain transformation using deep learning and autoencoder. Experimental results demonstrate the superior accuracy and efficiency of the method over the state-of-the-art.
Article
Computer Science, Information Systems
Yan Wang, Tongtong Su, Yusen Li, Jiuwen Cao, Gang Wang, Xiaoguang Liu
Summary: This paper proposes a lightweight network called DDistill-SR, which significantly improves the quality of super-resolution by capturing and reusing more helpful information. By using plug-in reparameterized dynamic units (RDU) and dynamic distillation fusion (DDF) modules, the network is able to achieve better performance while reducing parameters and computational overhead.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Review
Computer Science, Information Systems
Shutong Ye, Shengyu Zhao, Yaocong Hu, Chao Xie
Summary: This article reviews the single-image super-resolution (SISR) challenges from 2017 to 2022, focusing on different types of challenges, datasets used, evaluation methods, and outstanding network architectures. The article summarizes the methods widely used in recent years for SISR and suggests several potential research directions for the future.
Article
Computer Science, Artificial Intelligence
Zhechen Zhang, Weigang Lu, Shuo Chen, Fei Yang, Pan Jingchang
Summary: Recently, significant progress has been made in single image super-resolution (SISR) with the use of deep convolutional neural networks (CNN) and Generative Adversarial Networks (GAN). However, GAN-based methods suffer from lengthy and unstable convergence. To address these issues, this paper proposes a mechanism that incorporates boundary equilibrium in the image super-resolution network, allowing for balanced convergence of the generator and discriminator and improved visual quality of generated images. Additionally, the paper introduces an improved perceptual loss based on Learned Perceptual Image Patch Similarity (LPIPS), which outperforms traditional VGG-based perceptual loss in terms of acquiring better human visual effects. Experimental results demonstrate that the proposed method significantly enhances image super-resolution performance and achieves clearer details compared to state-of-the-art methods.
APPLIED INTELLIGENCE
(2023)
Review
Computer Science, Artificial Intelligence
Honggang Chen, Xiaohai He, Linbo Qing, Yuanyuan Wu, Chao Ren, Ray E. Sheriff, Ce Zhu
Summary: This article provides a comprehensive review of real-world single image super-resolution (RSISR), covering critical datasets, assessment metrics, and four major categories of RSISR methods. It compares representative RSISR methods on benchmark datasets in terms of reconstruction quality and computational efficiency, while also discussing challenges and promising research topics in RSISR.
INFORMATION FUSION
(2022)
Article
Computer Science, Artificial Intelligence
Zhiyong Huang, Wenbin Li, Jinxin Li, Dengwen Zhou
Summary: The proposed DPAN utilizes dual-path attention groups and dual skip connections to combine the advantages of residual and dense connections for better single image super resolution performance. The network can effectively utilize complementary contextual information extracted by different methods, and performs well on high-frequency information.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Yujie Dun, Zongyang Da, Shuai Yang, Yao Xue, Xueming Qian
Summary: The paper introduces single image super-resolution as an important computer vision task and proposes a new Kernel-Attended Residual Network (KARN) to address the shortcomings of traditional methods. KARN excels in feature fusion, feature representation, and extracts more advanced information, achieving significantly better performance than existing methods.
KNOWLEDGE-BASED SYSTEMS
(2021)
Review
Computer Science, Information Systems
Yoong Khang Ooi, Haidi Ibrahim
Summary: This paper reviews the development of image super-resolution technology, focusing on convolutional neural network-based algorithms. Different algorithms were compared in terms of datasets used, loss functions, evaluation metrics, etc., with the advantages and disadvantages of each upsampling module and design technique summarized.
Article
Computer Science, Artificial Intelligence
Muhammad Haris, Greg Shakhnarovich, Norimichi Ukita
Summary: Deep Back-Projection Networks (DBPN) utilize iterative up- and down-sampling layers to construct mutually-connected units, providing an error feedback mechanism and addressing the mutual dependencies between low- and high-resolution images. By extending this idea with parameter sharing and transition layers, DBPN achieves superior results across multiple data sets, particularly for large scaling factors like 8x.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2021)
Article
Computer Science, Artificial Intelligence
Tianyu Geng, Xiao-Yang Liu, Xiaodong Wang, Guiling Sun
Summary: This paper proposes a deep shearlet residual learning network (DSRLN) based on shearlet transform for estimating residual images, trained in the shearlet transform domain to provide optimal sparse approximation of cartoon-like images. To address the large statistical variation among shearlet coefficients, a dual-path training strategy and data weighting technique are introduced. Extensive evaluations show that DSRLN achieves close results in PSNR to state-of-the-art deep learning methods with fewer network parameters.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Computer Science, Information Systems
Meng Zhu, Wenjie Luo
Summary: The paper proposes a closed-loop residual attention network (CLRAN) to tackle the problem of single image super-resolution (SISR) and demonstrates its superiority over existing SISR methods in terms of performance and visual perception.
Article
Engineering, Electrical & Electronic
Alireza Esmaeilzehi, M. Omair Ahmad, M. N. S. Swamy
Summary: This paper proposes a new multi-domain residual block for image super resolution that combines spatial and spectral features to enhance the performance of light-weight super resolution networks.
IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING
(2021)
Article
Engineering, Electrical & Electronic
Rui Chen, Yan Zhang
Summary: This paper proposes a variational hybrid network with dynamic attention mechanisms for image super-resolution tasks. By utilizing a multi-scale variational encoder network and a curvature-domain loss, the method is able to generate more realistic and visually pleasing high-resolution images.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Sabina Umirzakova, Shabir Ahmad, Latif U. Khan, Taegkeun Whangbo
Summary: The integration of deep learning models and the IoT is revolutionizing healthcare and improving patient care. However, the utilization of low-resolution images generated by IoT devices introduces biases in deep learning models, impacting clinical decision-making. This survey highlights the need for accurate image restoration in medical imaging and emphasizes the role of developing precise super-resolution methods to enhance the quality of medical images and improve the performance of deep learning models in healthcare applications.
INFORMATION FUSION
(2024)
Article
Computer Science, Information Systems
Feng Huang, Zhifeng Wang, Jing Wu, Ying Shen, Liqiong Chen
Summary: A new residual triplet attention network (RTAN) is proposed in this study to enhance the learning ability of CNN in SISR, improving network stability and performance through innovative structures and modules.
Article
Computer Science, Artificial Intelligence
Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao, Chunjing Xu, Yixing Xu, Zhaohui Yang, Yiman Zhang, Dacheng Tao
Summary: Transformer, a deep neural network with a self-attention mechanism, has been initially used in natural language processing and is now gaining attention in computer vision tasks. Transformer-based models perform as well as or even better than convolutional and recurrent neural networks in various visual benchmarks. This paper reviews vision transformer models, categorizes them based on different tasks, and analyzes their advantages and disadvantages. The discussed categories include backbone network, high/mid-level vision, low-level vision, and video processing. Efficient methods for applying transformer in real device-based applications are also explored. The challenges and further research directions for vision transformers are discussed as well.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Stephen J. Maybank, Liu Liu, Dacheng Tao
Summary: This study defines a family of probability density functions on the unit hypersphere S-n and investigates their properties and parameter estimation methods. Various shapes of probability density functions can be obtained by adjusting the parameters. Experiments show that clustering algorithms based on Kullback-Leibler divergence can achieve good results even in high-dimensional scenarios.
JOURNAL OF MATHEMATICAL IMAGING AND VISION
(2023)
Article
Computer Science, Artificial Intelligence
Liu Liu, Ji Liu, Cho-Jui Hsieh, Dacheng Tao
Summary: This article discusses the composition problems of a specific form and proposes a stochastically controlled compositional gradient algorithm that significantly reduces computational cost. The proposed method improves composition algorithms under low target accuracy, as demonstrated through theoretical proofs and experiments.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Xu Yang, Cheng Deng, Zhiyuan Dang, Dacheng Tao
Summary: In this article, a novel multiview clustering model is proposed, which utilizes multiple autoencoder networks to embed multiview data into different latent spaces. A heterogeneous graph learning module is employed to adaptively fuse the latent representations, and intraview and interview collaborative learning are used to optimize the clustering results. Experimental results show that this method significantly outperforms other clustering approaches on multiple datasets.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Zhihao Cheng, Liu Liu, Aishan Liu, Hao Sun, Meng Fang, Dacheng Tao
Summary: Imitation learning from observation (LfO) is more preferable than imitation learning from demonstration (LfD) due to the non-necessity of expert actions. This article proves that LfO is almost equivalent to LfD in the deterministic robot environment and even in the robot environment with bounded randomness. Extensive experiments demonstrate that LfO achieves comparable performance to LfD. This suggests that LfO can be safely applied in practice without sacrificing performance compared to LfD.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Automation & Control Systems
Zongqi Liu, Xueguan Song, Chao Zhang, Yunsheng Ma, Dacheng Tao
Summary: This paper proposes a framework of randomized stacking with active learning for incremental multi-fidelity surrogate (MFS) modeling. It randomly projects the inputs of low-fidelity (LF) samples into different spaces and builds a series of LF regressors to capture the LF features. These base LF regressors are stacked to form the inputs of the subsequent incremental Gaussian process regression (iGPR) model for approximating the high-fidelity (HF) responses. The framework also adopts a query-by-committee (QBC)-based active learning method to incrementally update the current iGPR model.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Xikun Zhang, Dongjin Song, Dacheng Tao
Summary: Despite the progress in graph representation learning, little attention has been given to the continual learning scenario where new categories of nodes and their associated edges continuously emerge. Existing methods either ignore topological information or sacrifice stability for plasticity. To address this, the Hierarchical Prototype Networks (HPNs) extract abstract knowledge in the form of prototypes to represent expanded graphs. HPNs select relevant features and prototypes to adapt to new categories, maintaining performance over existing nodes. The experimental results show that HPNs outperform baseline techniques while consuming less memory.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Physics, Applied
Zeqiao Zhou, Yuxuan Du, Xinmei Tian, Dacheng Tao
Summary: The design of efficient combinatorial optimization algorithms is crucial in various fields such as logistics, finance, and chemistry. This article proposes the QAOA-in-QAOA (QAOA2) algorithm to address large-scale MaxCut problems using small quantum machines, by applying the divide-and-conquer heuristic. The performance of QAOA2 is proven to be competitive or better than classical algorithms when the node count is around 2000.
PHYSICAL REVIEW APPLIED
(2023)
Article
Computer Science, Artificial Intelligence
Shanshan Zhao, Mingming Gong, Haimei Zhao, Jing Zhang, Dacheng Tao
Summary: Recent studies have achieved promising results by jointly learning local feature detectors and descriptors. To overcome the lack of ground-truth keypoint supervision, previous methods have incorporated relevant knowledge about keypoint attributes into the network for enhanced model learning. This paper presents Deep Corner, an end-to-end deep network that combines a local similarity-based keypoint measure with a plain convolutional network, inspired by traditional corner detectors. The proposed method yields reliable keypoints, facilitate the learning of distinctive descriptors. Additionally, the paper introduces a multi-level U-Net architecture and a feature self-transformation operation to further improve keypoint localization and descriptor invariance.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2023)
Article
Automation & Control Systems
Xinyu Wu, Jinke Li, Liu Liu, Dacheng Tao
Summary: The lower limb power-assist exoskeletons are expected to assist paraplegic individuals in walking again. However, most exoskeletons only work in known environments, making it challenging to understand the user's intention and plan footstep sequences in unknown scenes. This study proposes a visual footstep planning system based on the Bezier curve, integrating Hololens and Realsense for environment understanding and user behavior intention recognition. Experimental results demonstrate that the planning time is reduced by 67.46% compared to traditional search algorithms, validating the effectiveness of the proposed system on a visual interaction platform.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Long Lan, Xiao Teng, Jing Zhang, Xiang Zhang, Dacheng Tao
Summary: In this study, an unsupervised person re-identification method is proposed, which has achieved great progress by training with pseudo labels. To purify the feature and label noise, multi-view features and the knowledge of a teacher model are utilized. Experimental results demonstrate the effectiveness of this approach for unsupervised person re-identification.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Han Zhao, Xu Yang, Cheng Deng, Dacheng Tao
Summary: In this study, we propose a structure-adaptive graph contrastive learning framework to capture potential discriminative relationships for improved graph representation learning.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Geochemistry & Geophysics
Di Wang, Qiming Zhang, Yufei Xu, Jing Zhang, Bo Du, Dacheng Tao, Liangpei Zhang
Summary: Large-scale vision models tailored to remote sensing tasks are proposed in this article, using Vision Transformers and a new rotated varied-size window attention mechanism. The experiments demonstrate the superior performance of the model in detection, classification, and segmentation tasks, as well as its advantages in terms of computational complexity and data efficiency.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Di Wang, Jing Zhang, Bo Du, Gui-Song Xia, Dacheng Tao
Summary: Deep learning has achieved great success in aerial image understanding for remote sensing research. However, most existing models are pretrained with ImageNet weights which hinder their fine-tuning performance on downstream aerial scene tasks due to domain gaps. This study empirically investigates RS pretraining on aerial images, training different networks from scratch using the MillionAID dataset to obtain pretrained backbones. Results show that RS pretraining enhances performance in scene recognition and RS-related semantics tasks, but task discrepancies still exist, highlighting the need for further research on large-scale pretraining datasets and effective methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Benjamin Kiefer, Matej Kristan, Janez Pers, Lojze Zust, Fabio Poiesi, Fabio Augusto de Alcantara Andrade, Alexandre Bernardino, Matthew Dawkins, Jenni Raitoharju, Yitong Quan, Adem Atmaca, Timon Hoefer, Qiming Zhang, Yufei Xu, Jing Zhang, Dacheng Tao, Lars Sommer, Raphael Spraul, Hangyue Zhao, Hongpu Zhang, Yanyun Zhao, Jan Lukas Augustin, Eui-Ik Jeon, Impyeong Lee, Luca Zedda, Andrea Loddo, Cecilia Di Ruberto, Sagar Verma, Siddharth Gupta, Shishir Muralidhara, Niharika Hegde, Daitao Xing, Nikolaos Evangeliou, Anthony Tzes, Vojtech Bartl, Jakub Spanhel, Adam Herout, Neelanjan Bhowmik, Toby P. Breckon, Shivanand Kundargi, Tejas Anvekar, Ramesh Ashok Tabib, Uma Mudengudi, Arpita Vats, Yang Song, Delong Liu, Yonglin Li, Shuman Li, Chenhao Tan, Long Lan, Vladimir Somers, Christophe De Vleeschouwer, Alexandre Alahi, Hsiang-Wei Huang, Cheng-Yen Yang, Jenq-Neng Hwang, Pyong-Kun Kim, Kwangju Kim, Kyoungoh Lee, Shuai Jiang, Haiwen Li, Ziqiang Zheng, Tuan-Anh Vu, Hai Nguyen-Truong, Sai-Kit Yeung, Zhuang Jia, Sophia Yang, Chih-Chung Hsu, Xiu-Yu Hou, Yu-An Jhang, Simon Yang, Mau-Tsuen Yang
Summary: The 1st Workshop on Maritime Computer Vision (MaCVi) 2023 focused on maritime computer vision for UAV and USV and organized subchallenges in areas such as object detection, tracking, obstacle segmentation, and detection. The report summarizes the main findings of the subchallenges and introduces a new benchmark called SeaDronesSee Object Detection v2. Over 130 submissions were evaluated and trends in the best-performing methodologies were assessed. The datasets, evaluation code, and leaderboard are publicly available.
2023 IEEE/CVF WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION WORKSHOPS (WACVW)
(2023)