P.(Pan) Wang
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
10 records found
1
MIRA
A Human-AI Co-Creation Agent for Self-reflection through Squiggle Game
Conference paper
(2026)
-
Yuting Jin, Tingting Wang, Dantong Qin, Zhibin Zhou, Mengkun Bi, Min Hua, Pan Wang
AI agents are increasingly explored for supporting reflection and well-being. However, we know little about how AI agents participate in reflective practices, particularly through co-creation. We presented MIRA, a co-creative AI agent that engages users in transforming abstract squiggles into concrete drawings while offering reflective feedback. Through a three-group comparative study, we examine how AI-mediated co-creation shapes reflective experience. We found that MIRA operates as a scaffold that introduces external reflective perspectives, helping users reinterpret experiences and surface emotions through visual expression. These findings highlight how AI agents can foster engaging co-creative experiences that encourage reflection among university students, while providing insights for designing AI-human co-creation that can extend to broader populations.
...
AI agents are increasingly explored for supporting reflection and well-being. However, we know little about how AI agents participate in reflective practices, particularly through co-creation. We presented MIRA, a co-creative AI agent that engages users in transforming abstract squiggles into concrete drawings while offering reflective feedback. Through a three-group comparative study, we examine how AI-mediated co-creation shapes reflective experience. We found that MIRA operates as a scaffold that introduces external reflective perspectives, helping users reinterpret experiences and surface emotions through visual expression. These findings highlight how AI agents can foster engaging co-creative experiences that encourage reflection among university students, while providing insights for designing AI-human co-creation that can extend to broader populations.
Understanding the neural basis of three-dimensional (3D) perception is a fundamental objective in cognitive neuroscience. Despite advances in decoding 2D visual stimuli from neural data, reconstructing high-fidelity 3D objects with detailed texture and geometry remains largely unexplored. In this work, we introduce NeuroSculptor3D, the first single-stage, end-to-end framework for reconstructing textured 3D shapes directly from brain activity. NeuroSculptor3D integrates a viewpoint-aware brain embedding module that captures fine-grained spatial variations across visual perspectives, and a hierarchical guidance mechanism that aligns brain-derived features with perceptual, semantic, and structural priors. Together, these components facilitate the generation of consistent multi-view embeddings, which are then decoded via TRELLIS to produce high-quality textured 3D reconstructions. Experiments on the fMRI-Shape dataset demonstrate that NeuroSculptor3D outperforms existing baselines across multiple settings, achieving significant improvements in both structural accuracy and semantic consistency. Code will be released to facilitate further research.
...
Understanding the neural basis of three-dimensional (3D) perception is a fundamental objective in cognitive neuroscience. Despite advances in decoding 2D visual stimuli from neural data, reconstructing high-fidelity 3D objects with detailed texture and geometry remains largely unexplored. In this work, we introduce NeuroSculptor3D, the first single-stage, end-to-end framework for reconstructing textured 3D shapes directly from brain activity. NeuroSculptor3D integrates a viewpoint-aware brain embedding module that captures fine-grained spatial variations across visual perspectives, and a hierarchical guidance mechanism that aligns brain-derived features with perceptual, semantic, and structural priors. Together, these components facilitate the generation of consistent multi-view embeddings, which are then decoded via TRELLIS to produce high-quality textured 3D reconstructions. Experiments on the fMRI-Shape dataset demonstrate that NeuroSculptor3D outperforms existing baselines across multiple settings, achieving significant improvements in both structural accuracy and semantic consistency. Code will be released to facilitate further research.
Journal article
(2026)
-
Iva Hristova, Florian Funk, Benjamin van Schaik, Fransje Clercx, Anna Ooms, Hanne Bosma, David Zaragozá Sabater, Amber Nonnekes, Pan Wang, Arjan Hillebrand
Optically pumped magnetometers (OPMs) have enabled wearable magnetoencephalography (MEG) systems, but achieving both scalability and precise sensor placement remains a challenge. Existing solutions, such as 3D-printed individualized helmets and multi-sized sensor arrays, either lack practicality for widespread use or require extensive manual adjustments. In this study, we present a novel, adjustable OPM-MEG helmet that accommodates a wide range of head sizes (corresponding to individuals aged 5 to 66+ years) while ensuring stable and accurate sensor positioning. The helmet features 80 sensor holders with a ratchet adjustment mechanism, allowing each OPM to be positioned on the subject's scalp without the need for multiple helmet sizes. A quick-release system ensures safety and ease of removal. Compared to existing solutions, the design proposed here eliminates the need for custom manufacturing, minimizes pre-scan adjustments, and avoids signal interference from electromagnetic tracking systems. Validation of sensor localization accuracy with a 3D-printed reference head and an electromagnetic co-registration approach showed a mean sensor positioning error of 2.1 mm and orientation error below 1°, demonstrating the effectiveness of our approach. This adjustable helmet provides a scalable and practical solution for OPM-MEG, facilitating broader adoption of wearable MEG in both research and clinical applications.
...
Optically pumped magnetometers (OPMs) have enabled wearable magnetoencephalography (MEG) systems, but achieving both scalability and precise sensor placement remains a challenge. Existing solutions, such as 3D-printed individualized helmets and multi-sized sensor arrays, either lack practicality for widespread use or require extensive manual adjustments. In this study, we present a novel, adjustable OPM-MEG helmet that accommodates a wide range of head sizes (corresponding to individuals aged 5 to 66+ years) while ensuring stable and accurate sensor positioning. The helmet features 80 sensor holders with a ratchet adjustment mechanism, allowing each OPM to be positioned on the subject's scalp without the need for multiple helmet sizes. A quick-release system ensures safety and ease of removal. Compared to existing solutions, the design proposed here eliminates the need for custom manufacturing, minimizes pre-scan adjustments, and avoids signal interference from electromagnetic tracking systems. Validation of sensor localization accuracy with a 3D-printed reference head and an electromagnetic co-registration approach showed a mean sensor positioning error of 2.1 mm and orientation error below 1°, demonstrating the effectiveness of our approach. This adjustable helmet provides a scalable and practical solution for OPM-MEG, facilitating broader adoption of wearable MEG in both research and clinical applications.
From thought to visual composition
A brain-driven visual blends technique for visual blending tasks
Visual blends is a design technique that combines elements from multiple images into harmonious compositions and has been increasingly explored as a means to support early-stage ideation in engineering design. However, existing blending workflows rely heavily on manual image selection and composition, making the process difficult, time-consuming, and skill-intensive for designers. In this work, we present a proof-of-concept brain-guided visual blends technique that integrates an EEG-to-image model to simplify the image acquisition process and a local image editing model to enable automated and controllable image composition. Our EEG-to-image model employs a two-stage training strategy, combining pretraining on large-scale unlabelled EEG data with fine-tuning in an EEG-conditioned diffusion model, achieving state-of-the-art performance in reconstructing visual stimuli. To support visual blending tasks, we incorporate a local editing model (Paint-by-Example) that generates coherent blends using user-provided masks, reference images, and backgrounds. A user study with 15 participants demonstrated that the model effectively supported the creation of visual blends that aligned with users' design vision, even without artistic skills. The results suggest that brain-guided blending can serve as a early-stage ideation interface in engineering design, helping designers iterate on mental concepts before formal modelling and evaluation.
...
Visual blends is a design technique that combines elements from multiple images into harmonious compositions and has been increasingly explored as a means to support early-stage ideation in engineering design. However, existing blending workflows rely heavily on manual image selection and composition, making the process difficult, time-consuming, and skill-intensive for designers. In this work, we present a proof-of-concept brain-guided visual blends technique that integrates an EEG-to-image model to simplify the image acquisition process and a local image editing model to enable automated and controllable image composition. Our EEG-to-image model employs a two-stage training strategy, combining pretraining on large-scale unlabelled EEG data with fine-tuning in an EEG-conditioned diffusion model, achieving state-of-the-art performance in reconstructing visual stimuli. To support visual blending tasks, we incorporate a local editing model (Paint-by-Example) that generates coherent blends using user-provided masks, reference images, and backgrounds. A user study with 15 participants demonstrated that the model effectively supported the creation of visual blends that aligned with users' design vision, even without artistic skills. The results suggest that brain-guided blending can serve as a early-stage ideation interface in engineering design, helping designers iterate on mental concepts before formal modelling and evaluation.
Journal article
(2025)
-
Pan Wang, Yash Khinvasara, Geesje Josine Creijghton, Tessa Scholing, Yihua Wang, Zhibin Zhou, Peter R.N. Childs, Yuan Yin
The emergence of large language models (LLMs) provides an opportunity for AI to operate as a co-ideation partner during the creative processes. However, designers currently lack a comprehensive methodology for engaging in co-ideation with LLMs, and there is a limited framework that describes the process of co-ideation between a designer and ChatGPT. This research thus aimed to explore how LLMs can act as codesigners and influence creative ideation processes of industrial designers and whether the ideation performance of a designer could be improved by employing the proposed framework for co-ideation with custom GPT. A survey was first conducted to detect how LLMs influenced the creative ideation processes of industrial designers and to understand the problems that designers face when using ChatGPT to ideate. Then, a framework which based on mapping content to guide the co-ideation between humans and custom GPT (named as Co-Ideator) was promoted. Finally, a design case study followed by a survey and an interview was conducted to evaluate the ideation performance of the custom GPT and framework compared with traditional ideation methods. Also, the effect of custom GPT on co-ideation was compared with a non-artificial intelligence (AI)-used condition. The findings indicated that if users employed co-ideation with custom GPT, the novelty and quality of ideation outperformed by using traditional ideation.
...
The emergence of large language models (LLMs) provides an opportunity for AI to operate as a co-ideation partner during the creative processes. However, designers currently lack a comprehensive methodology for engaging in co-ideation with LLMs, and there is a limited framework that describes the process of co-ideation between a designer and ChatGPT. This research thus aimed to explore how LLMs can act as codesigners and influence creative ideation processes of industrial designers and whether the ideation performance of a designer could be improved by employing the proposed framework for co-ideation with custom GPT. A survey was first conducted to detect how LLMs influenced the creative ideation processes of industrial designers and to understand the problems that designers face when using ChatGPT to ideate. Then, a framework which based on mapping content to guide the co-ideation between humans and custom GPT (named as Co-Ideator) was promoted. Finally, a design case study followed by a survey and an interview was conducted to evaluate the ideation performance of the custom GPT and framework compared with traditional ideation methods. Also, the effect of custom GPT on co-ideation was compared with a non-artificial intelligence (AI)-used condition. The findings indicated that if users employed co-ideation with custom GPT, the novelty and quality of ideation outperformed by using traditional ideation.
Towards stereoscopic vision
Attention-guided gaze estimation with EEG in 3D space
Since traditional gaze-tracking methods rely on line-of-sight estimation, spatial attention modeling from neural activity offers an alternative perspective to gaze estimation. This paper presents a proof-of-concept study on attention-guided gaze estimation with Electroencephalography (EEG), investigating whether brain signals can be leveraged to estimate attentional focus within a controlled 3D environment. We first conducted a preliminary survey to gather public opinions, revealing a generally positive attitude towards EEG-driven gaze tracking. Building on this insight, we collected an EEG dataset in VR, where participants engaged with stimuli presented at predefined spatial locations. We introduce a deep learning model that estimates the relative saliency of candidate positions, enabling gaze estimation through optimization within the learned representation. Our results demonstrate that attentional focus was successfully mapped in a 3D coordinate space from 5 participants, and low-frequency oscillations contributed more significantly to predictive performance. The model achieved robust accuracy in distinguishing gaze locations, highlighting the potential of EEG-based gaze estimation for attention tracking in 3D environments.
...
Since traditional gaze-tracking methods rely on line-of-sight estimation, spatial attention modeling from neural activity offers an alternative perspective to gaze estimation. This paper presents a proof-of-concept study on attention-guided gaze estimation with Electroencephalography (EEG), investigating whether brain signals can be leveraged to estimate attentional focus within a controlled 3D environment. We first conducted a preliminary survey to gather public opinions, revealing a generally positive attitude towards EEG-driven gaze tracking. Building on this insight, we collected an EEG dataset in VR, where participants engaged with stimuli presented at predefined spatial locations. We introduce a deep learning model that estimates the relative saliency of candidate positions, enabling gaze estimation through optimization within the learned representation. Our results demonstrate that attentional focus was successfully mapped in a 3D coordinate space from 5 participants, and low-frequency oscillations contributed more significantly to predictive performance. The model achieved robust accuracy in distinguishing gaze locations, highlighting the potential of EEG-based gaze estimation for attention tracking in 3D environments.
Enhanced sign evaluation with AI
A visual data-driven approach
The current evaluation of signs relies on quantitative comprehensibility testing. Such testing yields extensive findings about signs’ effectiveness. However, a shortcoming of comprehensibility testing is that it does not provide qualitative information relevant to sign modification and does not facilitate interactions between designers and users. This article advocates the use of visual data to evaluate signs by examining the similarities between signs and drawings produced by end users based on a sign referent given to them. A new evaluation index is developed to measure the extent to which a sign conforms to users’ mental images and to determine whether it should be redesigned. It is calculated by using the learned perceptual image patch similarity. To illustrate the modified approach, a study of safety signs is presented in the article. The article provides an example of how evaluation using visual data can be conducted.
...
The current evaluation of signs relies on quantitative comprehensibility testing. Such testing yields extensive findings about signs’ effectiveness. However, a shortcoming of comprehensibility testing is that it does not provide qualitative information relevant to sign modification and does not facilitate interactions between designers and users. This article advocates the use of visual data to evaluate signs by examining the similarities between signs and drawings produced by end users based on a sign referent given to them. A new evaluation index is developed to measure the extent to which a sign conforms to users’ mental images and to determine whether it should be redesigned. It is calculated by using the learned perceptual image patch similarity. To illustrate the modified approach, a study of safety signs is presented in the article. The article provides an example of how evaluation using visual data can be conducted.
Conference paper
(2025)
-
Pan Wang, Xun Zhang, Zhibin Zhou, Peter Childs, Kunpyo Lee, Maaike Kleinsmann, Stephen Jia Wang
Typeface design plays a vital role in graphic and communication design. Different typefaces are suitable for different contexts and can convey different emotions and messages. Typeface design still relies on skilled designers to create unique styles for specific needs. Recently, generative adversarial networks (GANs) have been applied to typeface generation, but these methods face challenges due to the high annotation requirements of typeface generation datasets, which are difficult to obtain. Furthermore, machine-generated typefaces often fail to meet designers’ specific requirements, as dataset annotations limit the diversity of the generated typefaces. In response to these limitations in current typeface generation models, we propose an alternative approach to the task. Instead of relying on dataset-provided annotations to define the typeface style vector, we introduce a transformer-based language model to learn the mapping between a typeface style description and the corresponding style vector. We evaluated the proposed model using both existing and newly created style descriptions. Results indicate that the model can generate high-quality, patent-free typefaces based on the input style descriptions provided by designers.
...
Typeface design plays a vital role in graphic and communication design. Different typefaces are suitable for different contexts and can convey different emotions and messages. Typeface design still relies on skilled designers to create unique styles for specific needs. Recently, generative adversarial networks (GANs) have been applied to typeface generation, but these methods face challenges due to the high annotation requirements of typeface generation datasets, which are difficult to obtain. Furthermore, machine-generated typefaces often fail to meet designers’ specific requirements, as dataset annotations limit the diversity of the generated typefaces. In response to these limitations in current typeface generation models, we propose an alternative approach to the task. Instead of relying on dataset-provided annotations to define the typeface style vector, we introduce a transformer-based language model to learn the mapping between a typeface style description and the corresponding style vector. We evaluated the proposed model using both existing and newly created style descriptions. Results indicate that the model can generate high-quality, patent-free typefaces based on the input style descriptions provided by designers.
Journal article
(2025)
-
Pan Wang, Xun Zhang, Liyan Wei, Peter Childs, Stephen Jia Wang, Yike Guo, Maaike Kleinsmann
Ideation is a critical step in the engineering design process, enabling designers to develop creative and innovative concepts and prototypes. Currently, the ideation workflow requires designers to generate new designs based on product requirements, heavily relying on their personal expertise and experience. To advance human-AI collaboration design and assist designers in the idea-generation process, this paper proposes an Object Combination Generative Adversarial Network (OC-GAN) for combinational creativity. The proposed method includes an image encoder module and a cross-domain object combination generator module. The image encoder module captures and encodes image structure information into latent space, while the cross-domain object combination generator module leverages GANs to combine object images based on user preferences, producing new design images. A design case study is used to evaluate the new ideation approach and reveal not only strong cross-domain concept combination capabilities but also improvement in designers' workflow and provision of novelty to the design case. Highlights An AI approach to improve the efficiency of idea generation in the design process. A case study evaluates its support for idea generation and design creativity. The OC-GAN is used for multi-domain object image combining tasks. Exemplifies the feasibility of human-AI collaboration design for enhancing creativity.
...
Ideation is a critical step in the engineering design process, enabling designers to develop creative and innovative concepts and prototypes. Currently, the ideation workflow requires designers to generate new designs based on product requirements, heavily relying on their personal expertise and experience. To advance human-AI collaboration design and assist designers in the idea-generation process, this paper proposes an Object Combination Generative Adversarial Network (OC-GAN) for combinational creativity. The proposed method includes an image encoder module and a cross-domain object combination generator module. The image encoder module captures and encodes image structure information into latent space, while the cross-domain object combination generator module leverages GANs to combine object images based on user preferences, producing new design images. A design case study is used to evaluate the new ideation approach and reveal not only strong cross-domain concept combination capabilities but also improvement in designers' workflow and provision of novelty to the design case. Highlights An AI approach to improve the efficiency of idea generation in the design process. A case study evaluates its support for idea generation and design creativity. The OC-GAN is used for multi-domain object image combining tasks. Exemplifies the feasibility of human-AI collaboration design for enhancing creativity.
Learning-based Artificial Intelligence Artwork
Methodology Taxonomy and Quality Evaluation
Journal article
(2024)
-
Qian Wang, Hong Ning Dai, Jinghua Yang, Cai Guo, Peter Childs, Maaike Kleinsmann, Yike Guo, Pan Wang
With the development of the theory and technology of computer science, machine or computer painting is increasingly being explored in the creation of art. Machine-made works are referred to as artificial intelligence (AI) artworks. Early methods of AI artwork generation have been classified as non-photorealistic rendering, and, latterly, neural style transfer methods have also been investigated. As technology advances, the variety of machine-generated artworks and the methods used to create them have proliferated. However, there is no unified and comprehensive system to classify and evaluate these works. To date, no work has generalized methods of creating AI artwork including learning-based methods for painting or drawing. Moreover, the taxonomy, evaluation, and development of AI artwork methods face many challenges. This article is motivated by these considerations. We first investigate current learning-based methods for making AI artworks and classify the methods according to art styles. Furthermore, we propose a consistent evaluation system for AI artworks and conduct a user study to evaluate the proposed system on different AI artworks. This evaluation system uses six criteria: beauty, color, texture, content detail, line, and style. The user study demonstrates that the six-dimensional evaluation index is effective for different types of AI artworks.
...
With the development of the theory and technology of computer science, machine or computer painting is increasingly being explored in the creation of art. Machine-made works are referred to as artificial intelligence (AI) artworks. Early methods of AI artwork generation have been classified as non-photorealistic rendering, and, latterly, neural style transfer methods have also been investigated. As technology advances, the variety of machine-generated artworks and the methods used to create them have proliferated. However, there is no unified and comprehensive system to classify and evaluate these works. To date, no work has generalized methods of creating AI artwork including learning-based methods for painting or drawing. Moreover, the taxonomy, evaluation, and development of AI artwork methods face many challenges. This article is motivated by these considerations. We first investigate current learning-based methods for making AI artworks and classify the methods according to art styles. Furthermore, we propose a consistent evaluation system for AI artworks and conduct a user study to evaluate the proposed system on different AI artworks. This evaluation system uses six criteria: beauty, color, texture, content detail, line, and style. The user study demonstrates that the six-dimensional evaluation index is effective for different types of AI artworks.