Article
Computer Science, Information Systems
Yuzhi Zhao, Lai-Man Po, Wing-Yin Yu, Yasar Abbas Ur Rehman, Mengyang Liu, Yujia Zhang, Weifeng Ou
Summary: We propose VCGAN, an improved approach to video colorization addressing the issues of temporal consistency and unification of colorization network and refinement network. VCGAN utilizes additional networks to enhance colorization quality and spatiotemporal consistency. Experimental results show that VCGAN produces higher-quality and temporally more consistent colorful videos than existing approaches.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Software Engineering
Min Shi, Jia-Qi Zhang, Shu-Yu Chen, Lin Gao, Yu-Kun Lai, Fang-Lue Zhang
Summary: This paper presents a deep architecture for automatically coloring line art videos with the same color style as the reference images. The proposed framework consists of a color transform network and a temporal refinement network. Experimental results show that our method achieves the best performance in line art video coloring.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2023)
Article
Mathematics
Xuefei Huang, Ka-Hou Chan, Wei Ke, Hao Sheng
Summary: This work proposes a parallel-based dense video captioning method that can address the mutual constraint between event proposals and captions. It introduces a deformable Transformer framework to reduce or eliminate manual threshold of hyperparameters. Experimental results show that the proposed method outperforms other methods in this area, providing competitive results on the ActivityNet Caption dataset.
Article
Computer Science, Artificial Intelligence
Yuxi Jin, Bin Sheng, Ping Li, C. L. Philip Chen
Summary: This automatic colorization method combines local and global features for coloring without user input and with short training time. The system uses local and global broad learning systems for obtaining and refining chrominance values, significantly reducing training time compared to traditional methods.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2021)
Article
Computer Science, Software Engineering
Michael Solah, Haikun Huang, Jiachuan Sheng, Tian Feng, Marc Pomplun, Lap-Fai Yu
Summary: One of the challenges in virtual scene design is invoking a specific mood in viewers, and we propose a novel approach to automatically adjust the colors of textures in a virtual indoor scene to match a target mood. Our approach was tested in different indoor scenes and showed efficacy in a user study using VR headsets.
IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS
(2022)
Article
Computer Science, Artificial Intelligence
Yuzhi Zhao, Lai-Man Po, Kangcheng Liu, Xuehui Wang, Wing-Yin Yu, Pengfei Xian, Yujia Zhang, Mengyang Liu
Summary: In this paper, a scribble-based video colorization network called SVCNet is proposed for colorizing monochrome videos based on different user-given color scribbles. It addresses common issues in the colorization area and utilizes two sequential sub-networks for precise colorization and temporal smoothing. The experimental results demonstrate that SVCNet produces higher-quality and more temporally consistent videos than other well-known video colorization approaches, as evaluated on DAVIS and Videvo benchmarks.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Hardware & Architecture
Xiangjun Zhao, Zhenlong Du
Summary: This paper proposes a text-driven video colorization method which incorporates a hybrid deep network for image object segmentation, foreground object colorization, background completion, text to image object linkage, and spatiotemporal preservation. The network, composed of a two-layer LSTM and mask RCNN, adopts a flexible training strategy to avoid the high cost of separate training while benefiting from text instruction, foreground object recognition, and location. Experiments show that the method outperforms state-of-the-art techniques in quantitative criteria and user studies confirm the preserved color naturalness, saturation, and richness in the colorized videos.
COMPUTERS & ELECTRICAL ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Yingxue Pang, Xin Jin, Jun Fu, Zhibo Chen
Summary: Deep learning techniques trained on large-scale datasets have made significant advancements in reference-based colorization. However, directly applying these methods to old photo colorization is challenging due to the lack of ground truth and domain gap. To tackle this problem, we propose a CNN-based algorithm called SFAC that trains on only two images for old photo colorization, overcoming the domain gap issue.
PATTERN RECOGNITION
(2024)
Article
Computer Science, Artificial Intelligence
Hai Wang, Xiaoyu Xiang, Yapeng Tian, Wenming Yang, Qingmin Liao
Summary: This research proposes a deformable attention network called STDAN for increasing the spatial-temporal resolution of low-resolution and low frame rate videos. It includes a LSTFI module for excavating content from neighboring input frames and a STDFA module for capturing and aggregating spatial and temporal contexts in dynamic video frames. Experimental results show that this approach outperforms existing STVSR methods.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Cheng Cheng, Hang Wang, Xiang Liao, Gang Cheng, Hongbin Sun
Summary: This paper proposes a infrared video colorization network CPNet, which aims to generate visually plausible and spatial-temporal consistent colorized videos. Experimental results demonstrate the superiority of CPNet in producing more authentic colorized videos compared to state-of-the-art methods.
COMPUTER VISION AND IMAGE UNDERSTANDING
(2023)
Article
Engineering, Electrical & Electronic
Minda Zhao, Qiang Ling
Summary: This paper proposes an adaptively meshed method to stabilize shaky videos, addressing issues in current video stabilization methods. By utilizing all feature trajectories and an adaptive blocking strategy, the method improves camera motion estimation performance and produces visually pleasing stabilization effects in challenging videos.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2021)
Article
Computer Science, Information Systems
Emanuele Marino, Fabio Bruno, Loris Barbieri, Maurizio Muzzupappa, Fotis Liarokapis
Summary: A major challenge in Augmented Reality (AR) is presenting augmented information effectively in various uncontrollable environmental conditions. The selection of colors plays a crucial role in blending virtual objects into the real environment, and different coloring strategies should be chosen based on the specific scenario to ensure optimal visibility and integration.
Article
Computer Science, Interdisciplinary Applications
Hakan Ezgi Kiziloz, Ayca Deniz
Summary: In this study, a robust framework for feature selection is built leveraging the multi-core nature of a regular PC. Multiple execution settings are facilitated through the use of two multiobjective selection algorithms, four initial population generation methods, and five machine learning techniques. Extensive experiments on 11 UCI benchmark datasets show remarkable improvement in terms of maximum accuracy.
COMPUTERS & INDUSTRIAL ENGINEERING
(2021)
Article
Computer Science, Artificial Intelligence
Zhongjie Zhuang, Jeng-Shyang Pan, Junbao Li, Shu-Chuan Chu
Summary: Arithmetic Optimization Algorithm (AOA) is a simple and easy to implement algorithm with few parameters. It utilizes the distribution behavior of arithmetic operators in mathematics. In this manuscript, AOA algorithm is converted into binary form with improved exploration using Multiplication Mathematical Optimizer Operator (MOO). Four families of transfer functions are used in the binary AOA (BAOA). Parallel mechanism is introduced to further enhance performance and proposed the Parallel Binary AOA (PBAOA) algorithm. Experimental results show that the proposed BAOA and PBAOA algorithms outperform classical and state-of-the-art algorithms in feature selection problems on low-dimensional and high-dimensional datasets.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Chemistry, Multidisciplinary
Carlos Ulloa, Dora M. Ballesteros, Diego Renza
Summary: This study presents two CNN-based models for classifying colorized and original images, evaluating the impact of hyperparameters. The transfer-learning-based model performs better in accuracy but has slower inference times compared to the custom model.
APPLIED SCIENCES-BASEL
(2021)
Article
Computer Science, Information Systems
Nan Jiang, Bin Sheng, Ping Li, Tong-Yee Lee
Summary: We introduce a new photographing guidance (PhotoHelper) that enhances the portrait photo quality for amateur photographers using deep feature retrieval and fusion. Our model combines empirical aesthetic rules, traditional machine learning algorithms, and deep neural networks to extract various features in color and space aspects. Through a series of processes, our approach provides users with professional reference photographs and automatically generates spatial composition guidance, resulting in significantly improved aesthetic quality of portrait photos.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Artificial Intelligence
Yu Zhou, Zhihua Chen, Ping Li, Haitao Song, C. L. Philip Chen, Bin Sheng
Summary: In this article, a novel end-to-end dehazing method called feedback spatial attention dehazing network (FSAD-Net) is proposed. FSAD-Net improves dehazing performance by utilizing feedback connections and attention mechanisms, and it shows superiority in both quantitative and qualitative experiments.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Engineering, Electrical & Electronic
Doolos Aibek Uulu, Rui Chen, Liang Chen, Ping Li, Hakan Bagci
Summary: This study solves a coupled system of volume integral and hydrodynamic equations to analyze electromagnetic scattering from nanostructures consisting of metallic and dielectric parts. The hydrodynamic equation relates the free electron polarization current to the electric flux, while the volume integral equation relates the electric flux and the free electron polarization current to the scattered electric field. An efficient two-level iterative solver is proposed to solve the matrix system generated by the expansions of the unknown electric flux and free electron polarization current. Numerical experiments demonstrate the accuracy, efficiency, and applicability of the proposed method.
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION
(2023)
Article
Computer Science, Information Systems
Xiao Lin, Shuzhou Sun, Wei Huang, Bin Sheng, Ping Li, David Dagan Feng
Summary: Recent transformer-based models, especially patch-based methods, have shown great potential in vision tasks. However, fixed-size patch division ignores the diversity of visual elements and hinders global attention information extraction. To address these issues, an Efficient Attention Pyramid Transformer (EAPT) is proposed, which includes Deformable Attention for non-fixed attention information and an Encode-Decode Communication module for complete global attention information. A specifically designed position encoding for vision transformers is also introduced. Extensive experiments and ablation studies demonstrate the effectiveness and key components of the proposed model.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Artificial Intelligence
Can Yang, Ping Li
Summary: Mobile game providers benefit from selling virtual items, and by utilizing multi-instance multi-label learning algorithm and semi-supervised learning, the recommendation performance of game props is improved, leading to a significant increase in game revenue, as confirmed by offline data sets and online game experimental results.
JOURNAL OF INTELLIGENT & FUZZY SYSTEMS
(2023)
Article
Materials Science, Characterization & Testing
Markus Wedekind, Susana Castillo, Marcus Magnor
Summary: This article introduces a method for correcting ring artifacts in computed tomography (CT) reconstruction. The method compensates for errors in the gain and offset values of each pixel and reduces blur by inferring information from neighboring pixels. Experimental results show that this method effectively mitigates the shortcomings of purely offset-based approaches and approaches using all projections, and can be efficiently implemented.
JOURNAL OF NONDESTRUCTIVE EVALUATION
(2023)
Article
Computer Science, Information Systems
Cai Guo, Qian Wang, Hong-Ning Dai, Ping Li
Summary: In this paper, a multi-stage feature-fusion dense network (MFFDNet) is proposed for motion deblurring, which improves the deblurring effect by reusing the extracted features and reducing computation costs. Experimental results show that MFFDNet outperforms state-of-the-art methods while maintaining a relatively small computing cost.
JOURNAL OF VISUAL COMMUNICATION AND IMAGE REPRESENTATION
(2023)
Article
Engineering, Electrical & Electronic
Xuan Zhang, Shi Min Liu, Ran Zhao, Xiao Chun Li, Jun Fa Mao, Li Jun Jiang, Ping Li
Summary: In this work, a wave-equation-based discontinuous Galerkin (DG) method hybridized with the Robin transmission condition (DG-RTC) is developed to solve frequency-domain electromagnetic (EM) problems. The proposed method discretizes the vector electric field wave equation in each subdomain and introduces numerical flux at the subdomain interfaces. An auxiliary equation based on the tangential continuity of EM fields is introduced to solve for the magnetic field H at the interfaces. A finite-element tearing and interconnecting (FETI)-like approach is used to reduce the computational cost. Several examples are presented to validate the proposed DG method.
IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES
(2023)
Article
Engineering, Biomedical
Jinyu Wen, Yang Li, Meie Fang, Lei Zhu, David Dagan Feng, Ping Li
Summary: The article introduces a novel wavelet convolution unit for image-oriented neural networks, combining wavelet analysis with convolution operator to extract deep abstract features efficiently. By fusing traditional convolution and wavelet decomposition, as well as incorporating multi-scale decompositions, it aims to improve classification accuracy.
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING
(2023)
Article
Engineering, Electrical & Electronic
Zi An Wang, Jun Fa Mao, Li Jun Jiang, Ping Li
Summary: This paper proposes a two-step source reconstruction method to address the localization and identification of unknown electromagnetic interference (EMI) sources in shielding enclosures. The actual EMI sources are initially modeled by equally distributed equivalent electric dipoles and the effects of surrounding environments are incorporated using the numerical Green's function. The parameters of equivalent dipoles are then updated to improve the precision of the equivalent model by introducing additional magnetic dipoles.
IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY
(2023)
Article
Engineering, Electrical & Electronic
Bo Li, Min Tang, Ping Li, Junfa Mao
Summary: In this article, an efficient approach named Laguerre-based layered finite element method (LB-LFEM) is presented for transient thermal analysis of integrated circuits (ICs) and packages with microchannel cooling. The LB-LFEM utilizes a marching-on-in-order scheme based on weighted Laguerre polynomials to eliminate the time variables and reduce the system matrix. By solving the Laguerre coefficients recursively, the computational efficiency is significantly improved. The validity and high efficiency of LB-LFEM are demonstrated by several examples.
IEEE JOURNAL ON MULTISCALE AND MULTIPHYSICS COMPUTATIONAL TECHNIQUES
(2023)
Article
Computer Science, Information Systems
Lei Zhu, Xiaoqiang Wang, Ping Li, Xin Yang, Qing Zhang, Weiming Wang, Carola-Bibiane Schonlieb, C. L. Philip Chen
Summary: RGB-D salient object detection aims to detect visually distinctive objects or regions from a pair of the RGB image and the depth image. In this work, we propose a self-supervised self-ensembling network (S-3 Net) for semi-supervised RGB-D salient object detection by leveraging the unlabeled data and exploring a self-supervised learning mechanism. Experimental results demonstrate that our network outperforms the state-of-the-art methods on seven widely-used benchmark datasets.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Artificial Intelligence
Hao Liu, Mang Ye, Yan Wang, Sanyuan Zhao, Ping Li, Jianbing Shen
Summary: This article presents a new adaptive metric distillation approach that significantly improves the backbone features of student networks and achieves better classification results. Previous knowledge distillation methods focused on transferring knowledge across classifier logits or feature structure, neglecting the excessive sample relations in the feature space. The proposed collaborative adaptive metric distillation (CAMD) optimizes the relationship between key pairs, adapts the metric for student embeddings using teacher embeddings as supervision, and employs a collaborative scheme for knowledge aggregation. Extensive experiments show that CAMD outperforms other cutting-edge distillers and sets a new state-of-the-art in both classification and retrieval tasks.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Zhihua Chen, Guhao Qiu, Ping Li, Lei Zhu, Xiaokang Yang, Bin Sheng
Summary: Recently, neural architecture search (NAS) has attracted attention from academia and industry. However, it remains a challenging problem due to the large search space and computational costs. In this study, a multi-teacher-guided NAS approach was proposed to improve the search efficiency and precision. Experimental results show that this approach achieves significant improvements in both accuracy and efficiency.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Education & Educational Research
Yuxi Jin, Ping Li, Wenxiao Wang, Suiyun Zhang, Di Lin, Chengjiu Yin
Summary: We have designed a generative adversarial network (GAN)-based pencil drawing learning system for art education on large image datasets. Students can learn how to draw pencil drawings for natural scenes by uploading an image and getting a pencil drawing example. Teachers can conveniently assign homework and understand the learning demands of students using the system.
INTERACTIVE LEARNING ENVIRONMENTS
(2023)
