Article
Computer Science, Artificial Intelligence
Yanlong Dong, Ying Zhang, Lin Ma, Zhi Wang, Jiebo Luo
Summary: This paper presents an unsupervised training approach for text-to-image synthesis, which generates pseudo image-text pair data based on visual concepts to initialize a GAN model. The proposed method is able to generate high-quality images for given sentences without the need for human-labeled data.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Information Systems
Bin Jiang, Yun Huang, Wei Huang, Chao Yang, Fangqiang Xu
Summary: This study proposes a Multi-scale Dual-modal Generative Networks (MD-GAN) for generating images from text descriptions. The method addresses two key issues in image generation: selectively aggregating channel information to adjust image texture and enhancing semantic consistency between text and images through the dual-modal modulation attention (DMA) and the multi-scale consistency discriminator (MCD).
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Bin Jiang, Weiyuan Zeng, Chao Yang, Renjun Wang, Bolin Zhang
Summary: In this paper, a novel Dual and Efficient Fusion Generative Adversarial Network (DE-GAN) is proposed to address the issues of limited diversity and high storage consumption in text-to-image synthesis. DE-GAN utilizes Dual Injection Blocks to balance the diversity and fidelity of generated images by injecting noise and text embeddings multiple times during the generation process. It also introduces an efficient condition channel attention module to capture correlations between text and image modalities with minimal storage overhead, enabling the model to adapt to resource-constrained applications. Comprehensive experiments demonstrate that DE-GAN efficiently generates more diverse and photo-realistic images compared to previous methods.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Review
Computer Science, Artificial Intelligence
Stanislav Frolov, Tobias Hinz, Federico Raue, Joern Hees, Andreas Dengel
Summary: Text-to-image synthesis has made significant progress in recent years but still faces challenges that require further research and improvement. Areas of focus include enhancing evaluation metrics and datasets, as well as improving model training and design.
Article
Computer Science, Theory & Methods
Meriem Guerar, Luca Verderame, Mauro Migliardi, Francesco Palmieri, Alessio Merlo
Summary: A recent study has shown that malicious bots generated a significant portion of website traffic in 2019, posing a serious threat to businesses. In order to combat these bots, introducing CAPTCHA tests has become a common defense mechanism. Therefore, understanding the effectiveness of different CAPTCHA schemes is crucial. This paper provides an overview of the current research in the field of CAPTCHA schemes and introduces a new classification. It also summarizes various attack methods and discusses the limitations of different CAPTCHA schemes.
ACM COMPUTING SURVEYS
(2022)
Article
Computer Science, Artificial Intelligence
Zhaorui Tan, Xi Yang, Zihan Ye, Qiufeng Wang, Yuyao Yan, Anh Nguyen, Kaizhu Huang
Summary: This paper tackles the challenge of generating high-quality images from text in visual-language understanding and introduces a novel text-image consistency metric, Semantic Similarity Distance (SSD). It also proposes Parallel Deep Fusion Generative Adversarial Networks (PDF-GAN) to mitigate inconsistent semantics and bridge the text-image semantic gap. Experimental results demonstrate that, guided by SSD, PDF-GAN significantly enhances the consistency between texts and images while preserving acceptable image quality.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Yue Ma, Li Liu, Huaxiang Zhang, Chunjing Wang, Zekang Wang
Summary: This paper proposes a novel generative adversarial network based on semantic consistency to generate semantically consistent and realistic images according to text descriptions. The method explores the semantic consistency between text and image for efficient cross-modal generation, and utilizes a generation network with hybrid attention to improve the authenticity of the generated images. Additionally, a semantic comparison module is introduced to compare the texts and generated images in the same semantic space through consistency refinement and information classification.
APPLIED INTELLIGENCE
(2023)
Article
Computer Science, Information Systems
Joo Yong Shim, Soyi Jung, Joongheon Kim, Jong-Kook Kim
Summary: This paper proposes an adaptive GAN selection scheme to generate new CAPTCHA images for enhanced security. The scheme focuses on maximizing image quality while ensuring system stability. Through performance evaluation, a trade-off between generation time and image quality is shown.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Information Systems
Fengling Mao, Bingpeng Ma, Hong Chang, Shiguang Shan, Xilin Chen
Summary: This study proposes an interstage cross-sample similarity distillation model based on a generative adversarial network (GAN) for learning efficient text-to-image synthesis. Experimental results show that the model generates visually pleasing images and achieves quantitatively comparable performance with state-of-the-art methods.
SCIENCE CHINA-INFORMATION SCIENCES
(2021)
Article
Computer Science, Information Systems
Hongchen Tan, Xiuping Liu, Baocai Yin, Xin Li
Summary: This paper proposes Cross-modal Semantic Matching Generative Adversarial Networks (CSM-GAN) to improve the semantic consistency between text description and synthesized image in fine-grained text-to-image generation. Two new modules, Text Encoder Module (TEM) and Textual-Visual Semantic Matching Module (TVSMM), are introduced to increase semantic consistency in the global semantic embedding space. Thorough experiments show the superiority of CSM-GAN over other representative state-of-the-art methods.
IEEE TRANSACTIONS ON MULTIMEDIA
(2022)
Article
Computer Science, Artificial Intelligence
Yong Xuan Tan, Chin Poo Lee, Mai Neo, Kian Ming Lim
Summary: Text-to-image synthesis is a technology that converts text descriptions into corresponding images and is widely used in various applications. However, this technology faces challenges such as visual realism, overconfidence, and training instability. To address these challenges, this paper proposes a self-supervised text-to-image synthesis method with enhancements including self-supervised learning, feature matching, L1 distance loss, and one-sided label smoothing. The proposed method generates images that are more diverse, visually realistic, and semantically consistent with the given text description.
PATTERN RECOGNITION LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Hongchen Tan, Xiuping Liu, Meng Liu, Baocai Yin, Xin Li
Summary: This paper presents a new framework called Knowledge-Transfer Generative Adversarial Network (KT-GAN) for fine-grained text-to-image generation. By introducing Alternate Attention-Transfer Mechanism (AATM) and Semantic Distillation Mechanism (SDM), the framework successfully bridges the cross-domain gap between text and image, achieving better text features and higher-quality images. Extensive experimental validation on two public datasets demonstrates that KT-GAN outperforms the baseline method significantly and achieves competitive results over various evaluation metrics.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2021)
Article
Computer Science, Artificial Intelligence
Dawei Zhu, Aditya Mogadala, Dietrich Klakow
Summary: The paper introduces a method of manipulating images using natural language descriptions, and designs TEA-cGAN to generate semantically manipulated images, including two different architectures for attending locations that need to be modified during generation and generating higher resolution images.
Article
Computer Science, Software Engineering
Chun Liu, Jingsong Hu, Hong Lin
Summary: This paper proposes SWF-GAN to synthesize images from descriptive text, which solves the problems of limited constraint of coarse-grained information and insufficient representational capacity of ordinary mask predictors. SWF-GAN designs a sentence-word fusion perceptual module to accurately generate the structure of the target object. The experiments show that SWF-GAN can generate clearer and more lively images.
COMPUTERS & GRAPHICS-UK
(2023)
Article
Computer Science, Artificial Intelligence
Tobias Hinz, Stefan Heinrich, Stefan Wermter
Summary: By introducing a new model and evaluation metric SOA, it provides a better evaluation of text-to-image models, ensuring that generated images match their captions and the user study confirmed this.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Chemistry, Analytical
Yi Du, Cuixia Ma, Chao Wu, Xiaowei Xu, Yike Guo, Yuanchun Zhou, Jianhui Li
Article
Engineering, Biomedical
Akara Supratak, Hao Dong, Chao Wu, Yike Guo
IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING
(2017)
Article
Computer Science, Theory & Methods
Hao Dong, Chao Wu, Zhen Wei, Yike Guo
IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY
(2018)
Article
Engineering, Biomedical
Hao Dong, Akara Supratak, Wei Pan, Chao Wu, Paul M. Matthews, Yike Guo
IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING
(2018)
Article
Neurosciences
Zhen Wei, Chao Wu, Xiaoyi Wang, Akara Supratak, Pan Wang, Yike Guo
FRONTIERS IN NEUROSCIENCE
(2018)
Article
Green & Sustainable Science & Technology
C. Pozo, P. Limleamthong, Y. Guo, T. Green, N. Shah, S. Acha, A. Sawas, C. Wu, M. Siegert, G. Guillen-Gosalbez
JOURNAL OF CLEANER PRODUCTION
(2019)
Article
Mathematical & Computational Biology
Xiaojie Huang, Jun Xiao, Chao Wu
Summary: This study proposes a model that uses deep neural networks to classify task states from fMRI data by simultaneously utilizing spatial and temporal information. By adding an attention mechanism to the recurrent network module, the model effectively highlights brain activation states at reaction moments, showing a high classification accuracy of 94.31% and the ability to distinguish brain states under different task stimuli.
COMPUTATIONAL INTELLIGENCE AND NEUROSCIENCE
(2021)
Article
Political Science
Chao Wu, Yixin Tu, Zexi Li, Jianxing Yu
Summary: The County Medical Community Reform, initiated in Anhui province and now widespread across China, has achieved noticeable results in governance, financing, creating resources, delivering services, and healthcare. During the COVID-19 outbreak, it has positively impacted the disease diagnosis and treatment in Zhejiang province’s primary health-care system. Lessons learned and recommendations for future development were summarized in an effort to optimize China’s primary health-care system.
JOURNAL OF CHINESE GOVERNANCE
(2021)
Article
Engineering, Electrical & Electronic
Bin Wang, Gang Li, Chao Wu, WeiShan Zhang, Jiehan Zhou, Ye Wei
Summary: This paper introduces a method called self-supervised federated domain adaptation (SFDA) to address the problem of distributed multi-source domain adaptation. SFDA effectively aggregates models from multiple source domains in each round of communication by proposing a multi-domain model generalization balance and a weighted strategy based on centroid similarity. It also tackles the domain shift in the target domain through self-supervised training and improves the accuracy of the model compared to classical federated adversarial domain adaptation algorithm.
EURASIP JOURNAL ON WIRELESS COMMUNICATIONS AND NETWORKING
(2022)
Article
Computer Science, Hardware & Architecture
Hongjing Huang, Zeke Wang, Jie Zhang, Zhenhao He, Chao Wu, Jun Xiao, Gustavo Alonso
Summary: This article discusses the characterization of High Bandwidth Memory (HBM) on an FPGA and the development of a benchmarking tool called Shuhai to evaluate its performance and usage. The study found that HBM can provide high memory bandwidth, but its usage significantly affects the achievable throughput. By comparing it with other types of memory, a better understanding of HBM's performance characteristics can be obtained.
IEEE TRANSACTIONS ON COMPUTERS
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Zhaoyang You, Xinya Wu, Kexuan Chen, Xinyi Liu, Chao Wu
Summary: In this study, a function was designed to recalculate Shapley Value, overcoming the issues caused by data replication and dataset partition, which led to an improvement in performance by about 50% when compared with the original index.
DATABASE AND EXPERT SYSTEMS APPLICATIONS, DEXA 2021, PT II
(2021)
Proceedings Paper
Computer Science, Information Systems
Chao Wu, Yi Cai, Mei Zhao, Songping Huang, Yike Guo
DATABASE SYSTEMS FOR ADVANCED APPLICATIONS, DASFAA 2016
(2016)
Proceedings Paper
Engineering, Biomedical
Shulin Yan, Xian Yang, Chao Wu, Yike Guo
2014 36TH ANNUAL INTERNATIONAL CONFERENCE OF THE IEEE ENGINEERING IN MEDICINE AND BIOLOGY SOCIETY (EMBC)
(2014)
Proceedings Paper
Computer Science, Artificial Intelligence
Shulin Yan, Lei Nie, Chao Wu, Yike Guo
BRAIN AND HEALTH INFORMATICS
(2013)