Hangwei Chen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Hangwei Chen has a total of 52 co-authors. Here are the top co-authors by number of shared publications:
- Feng Shao
Ningbo University
41 shared publications - Qiuping Jiang
School of Information Science and Engineering, Ningbo University, Ningbo 315211, China
27 shared publications - Xiongli Chai
Ningbo University
18 shared publications - Baoyang Mu
Faculty of Information Science and Engineering, Ningbo University, Ningbo, China
16 shared publications - Yo-Sung Ho
School of Information and Communications, Gwangju Institute of Science and Technology (GIST), Gwangju, South Korea
13 shared publications - Xiangchao Meng
Faculty of Electrical Engineering and Computer Science, Ningbo University, Ningbo, China
8 shared publications
Latest Publications
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (4) , 8070-8087
PU-T2M: Heterogeneous Knowledge Distilling Network From Transformer to Mamba for Lightweight Point Cloud Upsampling
Three-dimensional point clouds have become a core sensing modality in modern consumer electronics systems, particularly in intelligent vehicles equipped with low-power LiDAR and depth cameras. However, the point clouds captured by these consumer-grade sensors are often sparse and noisy due to complex environment and sensor constraints. Such degradation adversely affects downstream perception tasks, including real-time scene reconstruction, object understanding, and navigation stability. Point Cloud Upsampling (PCU) is therefore essential for restoring structural fidelity and enabling reliable three dimensions (3D) environment modeling in resource-constrained edge-to-cloud automotive systems. To overcome these issues, we propose a novel lightweight PCU architecture based on Heterogeneous Knowledge Distilling Network from Transformer to Mamba, termed PU-T2M. This workflow achieves state-of-the-art (SOTA) performance while being computationally efficient, making it suitable for edge devices with limited resources. Specifically, our network distills dark knowledge from Transformer to Mamba, enabling the model to learn long-range global context while preserving local geometric structure. We incorporate Mamba and knowledge distillation into the PCU task to achieve a more efficient and effective modeling framework. Specifically, we design a dual-alignment distillation loss that facilitates knowledge transfer while enforcing both explicit alignment in point distributions and implicit alignment in feature representations. In addition, we develop a Global Context Preserver module and a Local Geometry Preserver module to strengthen the Transformer’s global contextual modeling capability and enhance Mamba’s capacity for fine-grained local geometric learning. Extensive experiments conducted on four benchmark datasets demonstrate that PU-T2M attains state-of-the-art performance while maintaining a compact parameter scale and strong robustness. The proposed approach enables accurate and computationally efficient point cloud reconstruction, making it particularly suitable for industrial metrology and high-precision instrument-level quality inspection scenarios.
- 1 November 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Consumer Electronics
- Vol. 72 (4) , 8070-8087
PU-T2M: Heterogeneous Knowledge Distilling Network From Transformer to Mamba for Lightweight Point Cloud Upsampling
Three-dimensional point clouds have become a core sensing modality in modern consumer electronics systems, particularly in intelligent vehicles equipped with low-power LiDAR and depth cameras. However, the point clouds captured by these consumer-grade sensors are often sparse and noisy due to complex environment and sensor constraints. Such degradation adversely affects downstream perception tasks, including real-time scene reconstruction, object understanding, and navigation stability. Point Cloud Upsampling (PCU) is therefore essential for restoring structural fidelity and enabling reliable three dimensions (3D) environment modeling in resource-constrained edge-to-cloud automotive systems. To overcome these issues, we propose a novel lightweight PCU architecture based on Heterogeneous Knowledge Distilling Network from Transformer to Mamba, termed PU-T2M. This workflow achieves state-of-the-art (SOTA) performance while being computationally efficient, making it suitable for edge devices with limited resources. Specifically, our network distills dark knowledge from Transformer to Mamba, enabling the model to learn long-range global context while preserving local geometric structure. We incorporate Mamba and knowledge distillation into the PCU task to achieve a more efficient and effective modeling framework. Specifically, we design a dual-alignment distillation loss that facilitates knowledge transfer while enforcing both explicit alignment in point distributions and implicit alignment in feature representations. In addition, we develop a Global Context Preserver module and a Local Geometry Preserver module to strengthen the Transformer’s global contextual modeling capability and enhance Mamba’s capacity for fine-grained local geometric learning. Extensive experiments conducted on four benchmark datasets demonstrate that PU-T2M attains state-of-the-art performance while maintaining a compact parameter scale and strong robustness. The proposed approach enables accurate and computationally efficient point cloud reconstruction, making it particularly suitable for industrial metrology and high-precision instrument-level quality inspection scenarios.
- 10 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. PP (16) , 1
PU-TransMamba: A Hybrid Point Cloud Upsampling Framework with Detail-Aware Transformer and Spatially-Coherent Mamba
High-quality 3D point clouds are essential for high-fidelity perception in Internet of Things (IoT)-enabled intelligent systems. While Point Cloud Upsampling (PCU) is widely used to mitigate data sparsity, existing methods often struggle to balance the preservation of fine-grained local details with the maintenance of global topological consistency. Transformer-based approaches frequently suffer from excessive computational overhead and high-frequency detail loss, whereas emerging state space models like Mamba, despite their efficiency, inevitably sacrifice spatial coherence due to the 1D serialization of irregular 3D points. To address these critical bottlenecks, we introduce PU-TransMamba, a hybrid framework that synergistically leverages a Detail-Aware Transformer and a Spatially-Coherent Mamba. Each component is designed to resolve specific PCU limitations: a Complexity-Aware Bilateral Decoder is developed to adaptively recover sharp geometric edges by processing features across dual domains, while a Sequence-Aligned Mamba Encoder utilizes multiple spatial curvature descriptors to compensate for the spatial information loss inherent in serialization. Additionally, a Global Geometry Injector and a Local Neighbor Injector are designed to ensure structural integrity by infusing holistic skeletal priors and neighborhood context, respectively. To minimize feature discrepancies between the hybrid branches, we also propose a Self-Distillation Loss. Extensive experiments on five benchmark datasets demonstrate that PU-TransMamba outperforms state-of-the-art methods in both reconstruction accuracy and computational scalability. The results confirm its ability to recover intricate geometries, indicating significant potential for IoT-driven 3D perception and communication systems.
- 10 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. PP (16) , 1
PU-TransMamba: A Hybrid Point Cloud Upsampling Framework with Detail-Aware Transformer and Spatially-Coherent Mamba
High-quality 3D point clouds are essential for high-fidelity perception in Internet of Things (IoT)-enabled intelligent systems. While Point Cloud Upsampling (PCU) is widely used to mitigate data sparsity, existing methods often struggle to balance the preservation of fine-grained local details with the maintenance of global topological consistency. Transformer-based approaches frequently suffer from excessive computational overhead and high-frequency detail loss, whereas emerging state space models like Mamba, despite their efficiency, inevitably sacrifice spatial coherence due to the 1D serialization of irregular 3D points. To address these critical bottlenecks, we introduce PU-TransMamba, a hybrid framework that synergistically leverages a Detail-Aware Transformer and a Spatially-Coherent Mamba. Each component is designed to resolve specific PCU limitations: a Complexity-Aware Bilateral Decoder is developed to adaptively recover sharp geometric edges by processing features across dual domains, while a Sequence-Aligned Mamba Encoder utilizes multiple spatial curvature descriptors to compensate for the spatial information loss inherent in serialization. Additionally, a Global Geometry Injector and a Local Neighbor Injector are designed to ensure structural integrity by infusing holistic skeletal priors and neighborhood context, respectively. To minimize feature discrepancies between the hybrid branches, we also propose a Self-Distillation Loss. Extensive experiments on five benchmark datasets demonstrate that PU-TransMamba outperforms state-of-the-art methods in both reconstruction accuracy and computational scalability. The results confirm its ability to recover intricate geometries, indicating significant potential for IoT-driven 3D perception and communication systems.
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. PP (15209210) , 1-16
Toward Generalizable Blind Image Quality Assessment via Domain-Aware Hybrid Knowledge Learning
With the increasing diversity of image content formats and sources, recent years have witnessed a surge in blind image quality assessment (BIQA) methods tailored for diverse real-world scenarios. Although these task-specific BIQA models have achieved impressive performance, their generalization remains limited in broader evaluation contexts. Recently, mixed training has emerged as a promising solution to generalization challenges but remains underexplored. Current methods are often inflexibility or lack domain-aware modeling in the face of diverse quality degradation patterns. To address these issues, we propose a domain-aware hybrid knowledge generation network (DHKG-Net) through mixed training. Specifically, we first construct a hybrid knowledge generation module (HKGM), which adaptively generates domain-specific knowledge using knowledge banks to deal with diversity of quality degradation. Subsequently, we introduce a contrastive learning strategy to refine the feature distribution of domain-specific knowledge, thereby enhancing the effectiveness of domain-aware modeling. Finally, we design a knowledge-guided quality prediction branch to fuse the generated hybrid knowledge with initial visual features, significantly enhancing the model's adaptability to different application scenes. Extensive experiments are conducted on six datasets spanning three IQA tasks, and the results demonstrate that DHKG-Net substantially outperforms existing mixed training-based models in generalization capability. Furthermore, ablation studies and visual analyses validate the effectiveness and interpretability of the generated hybrid knowledge.
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. PP (15209210) , 1-16
Toward Generalizable Blind Image Quality Assessment via Domain-Aware Hybrid Knowledge Learning
With the increasing diversity of image content formats and sources, recent years have witnessed a surge in blind image quality assessment (BIQA) methods tailored for diverse real-world scenarios. Although these task-specific BIQA models have achieved impressive performance, their generalization remains limited in broader evaluation contexts. Recently, mixed training has emerged as a promising solution to generalization challenges but remains underexplored. Current methods are often inflexibility or lack domain-aware modeling in the face of diverse quality degradation patterns. To address these issues, we propose a domain-aware hybrid knowledge generation network (DHKG-Net) through mixed training. Specifically, we first construct a hybrid knowledge generation module (HKGM), which adaptively generates domain-specific knowledge using knowledge banks to deal with diversity of quality degradation. Subsequently, we introduce a contrastive learning strategy to refine the feature distribution of domain-specific knowledge, thereby enhancing the effectiveness of domain-aware modeling. Finally, we design a knowledge-guided quality prediction branch to fuse the generated hybrid knowledge with initial visual features, significantly enhancing the model's adaptability to different application scenes. Extensive experiments are conducted on six datasets spanning three IQA tasks, and the results demonstrate that DHKG-Net substantially outperforms existing mixed training-based models in generalization capability. Furthermore, ablation studies and visual analyses validate the effectiveness and interpretability of the generated hybrid knowledge.
- 17 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Broadcasting
- Vol. PP (00189316) , 1-13
Toward Multi-Dimensional Quality Assessment of AI-Generated Images via Prompt-Guided Relational Learning
The explosive growth of Artificial Intelligence Generated Images (AGIs) has created an urgent need for automated quality assessment. Unlike traditional image quality assessment, AGIs evaluation faces core challenges such as the absence of reference images, the need to integrate textual information, and the multi-dimensional nature of evaluation criteria. Existing methods typically rely on shared multimodal features for unified modeling, failing to effectively address two key issues: the differentiated extraction of dimension-specific features and the modeling of intrinsic relationships between dimensions. To address this, this paper proposes a Prompt-guided Relationship Learning Network (PRL-Net). Specifically, we first design a dimension-specific prompt attention (DPA) module, which introduces learnable prompts for each quality dimension to guide the model in adaptively focusing on dimension-related feature representations from shared features. Furthermore, we introduce a cross-dimensional graph relational attention (CGRA) module that explicitly models the complex dependencies between different quality dimensions (such as quality, consistency, and authenticity) through a dynamic graph neural network, enabling cross-dimensional feature interaction and information complementarity. Experimental results on multiple public benchmark datasets demonstrate that PRL-Net outperforms existing state-of-the-art methods across several evaluation metrics, validating its effectiveness.
- 17 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Broadcasting
- Vol. PP (00189316) , 1-13
Toward Multi-Dimensional Quality Assessment of AI-Generated Images via Prompt-Guided Relational Learning
The explosive growth of Artificial Intelligence Generated Images (AGIs) has created an urgent need for automated quality assessment. Unlike traditional image quality assessment, AGIs evaluation faces core challenges such as the absence of reference images, the need to integrate textual information, and the multi-dimensional nature of evaluation criteria. Existing methods typically rely on shared multimodal features for unified modeling, failing to effectively address two key issues: the differentiated extraction of dimension-specific features and the modeling of intrinsic relationships between dimensions. To address this, this paper proposes a Prompt-guided Relationship Learning Network (PRL-Net). Specifically, we first design a dimension-specific prompt attention (DPA) module, which introduces learnable prompts for each quality dimension to guide the model in adaptively focusing on dimension-related feature representations from shared features. Furthermore, we introduce a cross-dimensional graph relational attention (CGRA) module that explicitly models the complex dependencies between different quality dimensions (such as quality, consistency, and authenticity) through a dynamic graph neural network, enabling cross-dimensional feature interaction and information complementarity. Experimental results on multiple public benchmark datasets demonstrate that PRL-Net outperforms existing state-of-the-art methods across several evaluation metrics, validating its effectiveness.
- 24 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35 (10577149) , 3271-3282
Toward a Completely Blind Attacker for No-Reference Image Quality Assessment Models
No-reference image quality assessment (NR-IQA) models are critically vulnerable to adversarial attacks, posing significant risks to downstream vision systems. However, existing attack methods suffer from high computational costs, reliance on Mean Opinion Score (MOS) annotations, and poor cross-model transferability. To overcome these limitations, we propose Degrade-to-OverReconstruct (DOR), a novel prior knowledge-driven black-box attack framework operating in a “completely blind” manner, requiring neither MOS labels nor surrogate models, inducing significant prediction bias solely based on distortion statistics. Specifically, DOR generates universal adversarial examples by first applying mild degradation to preserve global structure and then employing aggressive over-reconstruction using a Residual Denoising Diffusion Model (RDDM) to adaptively disrupt intrinsic Natural Scene Statistics (NSS)—a shared foundation across NR-IQA models. Extensive experiments on synthetic (LIVE, TID2013) and authentic (CLIVE) datasets demonstrate DOR’s strong attack performance and superior transferability against leading NR-IQA models that cover diverse deep neural network architectures. Our work pioneers a diffusion model-based “completely blind” attack paradigm, offering a practical, MOS-free solution for adversarial robustness assessment of NR-IQA models in real-world deployments.
- 24 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 35 (10577149) , 3271-3282
Toward a Completely Blind Attacker for No-Reference Image Quality Assessment Models
No-reference image quality assessment (NR-IQA) models are critically vulnerable to adversarial attacks, posing significant risks to downstream vision systems. However, existing attack methods suffer from high computational costs, reliance on Mean Opinion Score (MOS) annotations, and poor cross-model transferability. To overcome these limitations, we propose Degrade-to-OverReconstruct (DOR), a novel prior knowledge-driven black-box attack framework operating in a “completely blind” manner, requiring neither MOS labels nor surrogate models, inducing significant prediction bias solely based on distortion statistics. Specifically, DOR generates universal adversarial examples by first applying mild degradation to preserve global structure and then employing aggressive over-reconstruction using a Residual Denoising Diffusion Model (RDDM) to adaptively disrupt intrinsic Natural Scene Statistics (NSS)—a shared foundation across NR-IQA models. Extensive experiments on synthetic (LIVE, TID2013) and authentic (CLIVE) datasets demonstrate DOR’s strong attack performance and superior transferability against leading NR-IQA models that cover diverse deep neural network architectures. Our work pioneers a diffusion model-based “completely blind” attack paradigm, offering a practical, MOS-free solution for adversarial robustness assessment of NR-IQA models in real-world deployments.
- 3 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Broadcasting
- Vol. PP (00189316) , 1-15
LV-BLiF: Harnessing Language-Vision Representation Learning for Blind Light Field Image Quality Assessment
Light Field Images (LFIs) capture both angular and spatial information, enabling immersive visual experiences for observers. However, during acquisition and processing, LFIs often suffer from various degradations, accompanied by intricate spatio-angular distortions. Consequently, objective LFI Quality Assessment (LFIQA) has garnered significant research interest. Existing blind LFIQA methods predominantly operate in a content-agnostic manner. They primarily model low-level distortions while neglecting the semantic context, a limitation that significantly diverges from the perception mechanisms of the Human Visual System (HVS). To bridge the gap between artifact analysis and scene understanding, we propose LV-BLiF, a novel dual-branch LFIQA network based on language-vision representation learning. Firstly, the textual prompts-assisted branch utilizes a pre-trained quality-aware Large Multimodal Model (LMM) to extract deep semantic representations, which are transformed into semantic context vectors sensitive to spatial-angular quality variations via pseudo-label guided learning strategy. Subsequently, semantic quality scores are generated based on this high-level information. Secondly, the visual cues-assisted branch employs disentangling convolutions to decompose light field features into spatial, angular, and epipolar plane image subspaces. These features are processed by Light Field Residual Blocks (LFRBs) and adaptively integrated via a Mixture-of-Experts-like (MoE-like) gating mechanism to predict visual quality scores. Finally, the semantic and visual scores are fused through a learnable adaptive gating mechanism to derive the overall LFI quality score. Extensive experiments on benchmark datasets, including NBU, SHU, and Win5-LID, demonstrate that LV-BLiF significantly outperforms state-of-the-art methods and exhibits high consistency with human perception. The proposed metric will be publicly available at https://github.com/shuaibilx/LV-BLiF
- 3 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Broadcasting
- Vol. PP (00189316) , 1-15
LV-BLiF: Harnessing Language-Vision Representation Learning for Blind Light Field Image Quality Assessment
Light Field Images (LFIs) capture both angular and spatial information, enabling immersive visual experiences for observers. However, during acquisition and processing, LFIs often suffer from various degradations, accompanied by intricate spatio-angular distortions. Consequently, objective LFI Quality Assessment (LFIQA) has garnered significant research interest. Existing blind LFIQA methods predominantly operate in a content-agnostic manner. They primarily model low-level distortions while neglecting the semantic context, a limitation that significantly diverges from the perception mechanisms of the Human Visual System (HVS). To bridge the gap between artifact analysis and scene understanding, we propose LV-BLiF, a novel dual-branch LFIQA network based on language-vision representation learning. Firstly, the textual prompts-assisted branch utilizes a pre-trained quality-aware Large Multimodal Model (LMM) to extract deep semantic representations, which are transformed into semantic context vectors sensitive to spatial-angular quality variations via pseudo-label guided learning strategy. Subsequently, semantic quality scores are generated based on this high-level information. Secondly, the visual cues-assisted branch employs disentangling convolutions to decompose light field features into spatial, angular, and epipolar plane image subspaces. These features are processed by Light Field Residual Blocks (LFRBs) and adaptively integrated via a Mixture-of-Experts-like (MoE-like) gating mechanism to predict visual quality scores. Finally, the semantic and visual scores are fused through a learnable adaptive gating mechanism to derive the overall LFI quality score. Extensive experiments on benchmark datasets, including NBU, SHU, and Win5-LID, demonstrate that LV-BLiF significantly outperforms state-of-the-art methods and exhibits high consistency with human perception. The proposed metric will be publicly available at https://github.com/shuaibilx/LV-BLiF
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (7) , 12765-12779
SGNet: A Structure-Guided Lightweight Network for VDT Salient Object Detection
Visible–depth–thermal (VDT) salient object detection (SOD) aims to jointly exploit RGB, depth and thermal cues to segment the most visually significant regions. However, most existing VDT SOD models are heavy in parameters and computational cost, limiting their deployment on real-world and edge devices. To tackle this, we propose structure-guided network (SGNet), a structure-guided lightweight network for efficient VDT SOD. Specifically, we design a lightweight tri-modal fusion module (TFM) to integrate three modalities at the semantic level, and a shared structure extraction module (SSEM) to extract common structural information from depth and thermal modalities. A structure refine module (SRM) further injects the extracted structure into the deepest semantic features, while a multiscale feature refinement module (MFRM) progressively decodes multilevel features under deep supervision to produce saliency maps with clear boundaries. Benefiting from these modules, SGNet achieves competitive performance on the VDT2048 benchmark with only 5.51M parameters and a real-time speed of 120 frames/s (FPS) at $320\times 320$ resolution, surpassing state-of-the-art methods while remaining deployment-friendly.
- 12 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (7) , 12765-12779
SGNet: A Structure-Guided Lightweight Network for VDT Salient Object Detection
Visible–depth–thermal (VDT) salient object detection (SOD) aims to jointly exploit RGB, depth and thermal cues to segment the most visually significant regions. However, most existing VDT SOD models are heavy in parameters and computational cost, limiting their deployment on real-world and edge devices. To tackle this, we propose structure-guided network (SGNet), a structure-guided lightweight network for efficient VDT SOD. Specifically, we design a lightweight tri-modal fusion module (TFM) to integrate three modalities at the semantic level, and a shared structure extraction module (SSEM) to extract common structural information from depth and thermal modalities. A structure refine module (SRM) further injects the extracted structure into the deepest semantic features, while a multiscale feature refinement module (MFRM) progressively decodes multilevel features under deep supervision to produce saliency maps with clear boundaries. Benefiting from these modules, SGNet achieves competitive performance on the VDT2048 benchmark with only 5.51M parameters and a real-time speed of 120 frames/s (FPS) at $320\times 320$ resolution, surpassing state-of-the-art methods while remaining deployment-friendly.
- 6 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28 (15209210) , 57-70
Quality Evaluation of AI-Generated Images: Subjective Study and Objective Methodology
In recent years, AI-Generated Images (AIGIs) have attracted significant attention and shown great potential in various applications, including entertainment, advertisement, education, and product design. Driven by this trend, various Text-to-Image (T2I) models are developed. However, the quality of AIGIs produced by these models varies widely, with many low-quality images failing to meet human aesthetic standards. Consequently, research into both subjective and objective Image Quality Assessment (IQA) methods for AIGIs is crucial. In this paper, we introduce a dataset called AIGI-IQAD, designed to enhance our understanding of human aesthetic preferences for AIGIs. The dataset contains 2,880 AIGIs generated by 8 T2I models using 360 deliberately designed text prompts. Further, we conducted subjective experiments to gather ratings from both aesthetic quality and text-image consistency. Building on this dataset, we propose a model named Question-guided Multimodal Interaction Network (QMI-Net) for evaluating AIGIs. QMI-Net assesses human preferences for AIGIs by focusing on both aesthetic quality and text-image consistency. Specifically, QMI-Net uses a question-answering approach to guide Multimodal Large Language Models (MLLMs) in generating detailed aesthetic and similarity information. The Visual and Aesthetic Feature Fusion Module (VAFFM) then fuses the aesthetic features with the visual features extracted by Contrastive Language-Image Pre-training (CLIP) to obtain more comprehensive aesthetic quality features. Comprehensive experiments demonstrate that state-of-the-art performance is achieved by QMI-Net on our AIGI-IQAD and three other public datasets.The AIGI-IQAD datasets and QMI-Net will be released at https://github.com/ctxya1207/QMI-Net.
- 6 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Multimedia
- Vol. 28 (15209210) , 57-70
Quality Evaluation of AI-Generated Images: Subjective Study and Objective Methodology
In recent years, AI-Generated Images (AIGIs) have attracted significant attention and shown great potential in various applications, including entertainment, advertisement, education, and product design. Driven by this trend, various Text-to-Image (T2I) models are developed. However, the quality of AIGIs produced by these models varies widely, with many low-quality images failing to meet human aesthetic standards. Consequently, research into both subjective and objective Image Quality Assessment (IQA) methods for AIGIs is crucial. In this paper, we introduce a dataset called AIGI-IQAD, designed to enhance our understanding of human aesthetic preferences for AIGIs. The dataset contains 2,880 AIGIs generated by 8 T2I models using 360 deliberately designed text prompts. Further, we conducted subjective experiments to gather ratings from both aesthetic quality and text-image consistency. Building on this dataset, we propose a model named Question-guided Multimodal Interaction Network (QMI-Net) for evaluating AIGIs. QMI-Net assesses human preferences for AIGIs by focusing on both aesthetic quality and text-image consistency. Specifically, QMI-Net uses a question-answering approach to guide Multimodal Large Language Models (MLLMs) in generating detailed aesthetic and similarity information. The Visual and Aesthetic Feature Fusion Module (VAFFM) then fuses the aesthetic features with the visual features extracted by Contrastive Language-Image Pre-training (CLIP) to obtain more comprehensive aesthetic quality features. Comprehensive experiments demonstrate that state-of-the-art performance is achieved by QMI-Net on our AIGI-IQAD and three other public datasets.The AIGI-IQAD datasets and QMI-Net will be released at https://github.com/ctxya1207/QMI-Net.