A5024109802- Generative Adversarial Networks and Image Synthesis
- Aviation Industry Analysis and Trends
- Vehicle emissions and performance
- Aesthetic Perception and Analysis
- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Transportation Planning and Optimization
- Digital Games and Media
- Artificial Intelligence in Games
- Art History and Market Analysis
- Gender Roles and Identity Studies
- Radiative Heat Transfer Studies
- Handwritten Text Recognition Techniques
- Human Pose and Action Recognition
- Robotics and Sensor-Based Localization
- Gender Studies in Language
- Transport and Economic Policies
- Manufacturing Process and Optimization
- Educational Games and Gamification
- Media, Gender, and Advertising
- Advanced Image Processing Techniques
- Robotic Path Planning Algorithms
- Music and Audio Processing
- Advanced Manufacturing and Logistics Optimization
- Cinema and Media Studies
Osaka University
2023-2025
Shandong Academy of Sciences
2024
Qilu University of Technology
2024
Tsinghua University
2023
China Railway Group (China)
2023
China Railway 18th Bureau Group Corporation
2023
Beijing Institute of Technology
2019
The increasing tendency to collect large and uncurated datasets train vision-and-language models has raised concerns about fair representations. It is known that even small but manually annotated datasets, such as MSCOCO, are affected by societal bias. This problem, far from being solved, may be getting worse with data crawled the Internet without much control. In addition, lack of tools analyze bias in big collections images makes addressing problem extremely challenging. Our first...
Text-to-image models have demonstrated remarkable capabilities in producing high-fidelity images from natural language prompts. The widespread application and increasing accessibility of pioneering models, such as Stable Diffusion, gained significant attention regarding the impact generated on representations downstream tasks. Concurrently, ethical considerations text-to-image generation emerged especially gender bias. This paper presents three projects that explore generative their first...
The duality of content and style is inherent to the nature art. For humans, these two elements are clearly different: refers objects concepts in piece art, way it expressed. This poses an important challenge for computer vision. visual appearance modulated by that may reflect author's emotions, social trends, artistic movement, etc., their deep comprehension undoubtfully requires handle both. A promising step towards a general paradigm art analysis disentangle style, whereas relying on human...
Social biases in generative models have gained increasing attention. This paper proposes an automatic evaluation protocol for text-to-image generation, examining how gender bias originates and perpetuates the generation process of Stable Diffusion. Using triplet prompts that vary by indicators, we trace presentations at several stages explore dependencies between images. Our findings reveal persists throughout all internal generating manifests entire For instance, differences object...
Abstract China built the longest high-speed railway system by consuming massive construction materials. However, characterization material metabolism in HSR remains less explored. Here we conducted a bottom-up study and revealed stocks, flows, greenhouse gas emissions from 2008 to 2035 China’s railway. We show that stocks temporally amount 0.6 gigatons 2010 3.7 2020, dominated aggregate cement. Spatially, stock distribution gaps across Chinese provinces are becoming more narrowed. Material...
Several studies have raised awareness about social biases in image generative models, demonstrating their predisposition towards stereotypes and imbalances. This paper contributes to this growing body of research by introducing an evaluation protocol that analyzes the impact gender indicators at every step generation process on Stable Diffusion images. Leveraging insights from prior work, we explore how not only affect presentation but also representation objects layouts within generated Our...
In order to reduce the labor and material of library management contribute development unmanned library, an intelligent mobile robot for is proposed. The consists a platform, three-DOF rise-and-fall robotic arm multisource image recognition information fusion system. Besides recognizing data position book, also able conduct book grasp shelving in task. To adapt existing environment enlarge its move area, efficient autonomous elevator button system based on neural networks designed, which...
With the rapid development of information age, images and animations have become mainstream way product display at present. At same time, image processing technology makes a reality. Through processing, proportion subject to whole is calculated, sample its are obtained. A more accurate psychological model users' needs obtained by using joint experiment eye movement EEG, influence ratio on transmission determined. The purpose this study provide reference suggestions for design publicity...
In this companion paper, we provide the artifacts of GOYA model for disentangling content and style in art paintings, as presented at ICMR2023. The scripts are written Python.
The content-style duality is a fundamental element in art. These two dimensions can be easily differentiated by humans: content refers to the objects and concepts an artwork, style way it looks. Yet, we have not found fully capture this with visual representations. While transfer captures appearance of single fails generalize larger sets. Similarly, supervised classification-based methods are impractical since perception lies on spectrum categorical labels. We thus present
The rapid development of text-to-image generation has brought rising ethical considerations, especially regarding gender bias. Given a text prompt as input, models generate images according to the prompt. Pioneering such Stable Diffusion and DALL-E 2 have demonstrated remarkable capabilities in producing high-fidelity from natural language prompts. However, these often exhibit bias, studied by tendency generating man prompts "a photo software developer". widespread application increasing...
Recent studies have highlighted biases in generative models, shedding light on their predisposition towards gender-based stereotypes and imbalances. This paper contributes to this growing body of research by introducing an evaluation protocol designed automatically analyze the impact gender indicators Stable Diffusion images. Leveraging insights from prior work, we explore how not only affect presentation but also representation objects layouts within generated Our findings include existence...
Abstract China built the longest high-speed rail (HSR) network by consuming massive construction materials. However, characterization of HSR material metabolism and associated emissions at national level remains less explored. Here, we revealed life-cycle GHG from 2008 to 2035 in China’s network. We show that stocks temporally amount 0.6 gigatons 2010 3.7 2020, dominated aggregate cement rather than steel. Spatially, distributions across provinces are becoming balanced. Growing speed...
The increasing tendency to collect large and uncurated datasets train vision-and-language models has raised concerns about fair representations. It is known that even small but manually annotated datasets, such as MSCOCO, are affected by societal bias. This problem, far from being solved, may be getting worse with data crawled the Internet without much control. In addition, lack of tools analyze bias in big collections images makes addressing problem extremely challenging. Our first...