Yuchun Fang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yuchun Fang has a total of 102 co-authors. Here are the top co-authors by number of shared publications:
Wei Zhang
7 shared publicationsYifan Li
Shanghai University
6 shared publicationsQiulong Yuan
Shanghai University
4 shared publications- Shiquan Lin
Shanghai University
4 shared publications Shouyong Pan
4 shared publicationsSirui Cai
School of Computer Engineering and Science, Shanghai University, Shanghai, China
4 shared publications
Latest Publications
- Open Access
- 11 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 22 (11) , 1-23
Monocular Interacting-Hand Reconstruction with Multimodal Context Fusion and Spatial Attention
3D hand reconstruction from monocular RGB images has attracted increasing attention due to its low cost and ease of deployment. However, accurately reconstructing interacting hands remains challenging, primarily because mutual occlusion leads to missing visual evidence, while image cropping weakens global spatial consistency. To address these issues, we propose the Context-Aware Interacting Hand Reconstruction Network (CANet), a monocular interacting-hand reconstruction framework that integrates multimodal context fusion with structured spatial attention. Specifically, CANet leverages estimated depth maps, edge contours, and center heatmaps to retain global contextual cues and guide reconstruction in occluded regions. A hierarchical spatial attention module further enhances spatial consistency by separately modeling intra-hand structural dependencies and inter-hand interactions, enabling more coherent reasoning about complex hand poses. Experiments across multiple public benchmarks demonstrate that CANet consistently improves the accuracy of both hand pose estimation and mesh reconstruction.
- 11 November 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Multimedia Computing, Communications, and Applications
- Vol. 22 (11) , 1-23
Open AccessMonocular Interacting-Hand Reconstruction with Multimodal Context Fusion and Spatial Attention
3D hand reconstruction from monocular RGB images has attracted increasing attention due to its low cost and ease of deployment. However, accurately reconstructing interacting hands remains challenging, primarily because mutual occlusion leads to missing visual evidence, while image cropping weakens global spatial consistency. To address these issues, we propose the Context-Aware Interacting Hand Reconstruction Network (CANet), a monocular interacting-hand reconstruction framework that integrates multimodal context fusion with structured spatial attention. Specifically, CANet leverages estimated depth maps, edge contours, and center heatmaps to retain global contextual cues and guide reconstruction in occluded regions. A hierarchical spatial attention module further enhances spatial consistency by separately modeling intra-hand structural dependencies and inter-hand interactions, enabling more coherent reasoning about complex hand poses. Experiments across multiple public benchmarks demonstrate that CANet consistently improves the accuracy of both hand pose estimation and mesh reconstruction.
- 16 October 2026
- Published by Elsevier in Information Processing & Management
- Vol. 64 (2)
Generative semantic-enhanced vision-language pre-training for sign language translation
- 16 October 2026
- Published by Elsevier in Information Processing & Management
- Vol. 64 (2)
Generative semantic-enhanced vision-language pre-training for sign language translation
- Open Access
- 22 July 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Weighted estimation of multiple regions in CT-derived images of the pubic symphysis for age estimation using a deep learning framework with R-Mixup augmentation
- 22 July 2026
- Published by Springer Nature in Scientific Reports
- Vol. 16 (1)
Open AccessWeighted estimation of multiple regions in CT-derived images of the pubic symphysis for age estimation using a deep learning framework with R-Mixup augmentation
- 4 May 2026
- Published by Elsevier in Information Processing & Management
- Vol. 63 (7)
Deciphering oracle bone characters: Pictographic captioning and reasoning for morphological recognition
- 4 May 2026
- Published by Elsevier in Information Processing & Management
- Vol. 63 (7)
Deciphering oracle bone characters: Pictographic captioning and reasoning for morphological recognition
- 27 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33 (10709908) , 1781-1785
Diffusion-Driven Forgery Detection: Distilling Latent Features for Generalized Image Forensics
Generalized detection of diffusion-generated forgeries across unseen generators is essential for maintaining visual authenticity in the era of generative media. Recent studies incorporate pretrained diffusion models into detection frameworks, primarily capturing forgery cues via image-level reconstruction through the complete diffusion-denoising process. However, such schemes lack an explicit separation between the semantic priors encoded in diffusion representations and the generator-specific noise trajectories introduced during synthesis, causing detectors to overfit to artifacts from specific training generators. To address this limitation, we propose a two-stage diffusion-guided framework that exploits latent-space priors within a pretrained denoising backbone and decouples the detection process into authentic manifold modeling and forgery discrimination stages. In the authentic manifold modeling phase, stable real-image representations are distilled from diffusion features using only authentic data to establish a generator-agnostic semantic space. During forgery discrimination, forged samples are incorporated and multi-layer residual modeling is designed to capture representation inconsistencies relative to the distilled real-image manifold. Extensive experiments conducted on multiple diffusion forgery benchmarks demonstrate that the proposed approach achieves consistent cross-generator generalization performance.
- 27 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 33 (10709908) , 1781-1785
Diffusion-Driven Forgery Detection: Distilling Latent Features for Generalized Image Forensics
Generalized detection of diffusion-generated forgeries across unseen generators is essential for maintaining visual authenticity in the era of generative media. Recent studies incorporate pretrained diffusion models into detection frameworks, primarily capturing forgery cues via image-level reconstruction through the complete diffusion-denoising process. However, such schemes lack an explicit separation between the semantic priors encoded in diffusion representations and the generator-specific noise trajectories introduced during synthesis, causing detectors to overfit to artifacts from specific training generators. To address this limitation, we propose a two-stage diffusion-guided framework that exploits latent-space priors within a pretrained denoising backbone and decouples the detection process into authentic manifold modeling and forgery discrimination stages. In the authentic manifold modeling phase, stable real-image representations are distilled from diffusion features using only authentic data to establish a generator-agnostic semantic space. During forgery discrimination, forged samples are incorporated and multi-layer residual modeling is designed to capture representation inconsistencies relative to the distilled real-image manifold. Extensive experiments conducted on multiple diffusion forgery benchmarks demonstrate that the proposed approach achieves consistent cross-generator generalization performance.
- 23 January 2026
- Published by Springer Nature
- p. 226-240
DiscoIB: Disentangled Subject Customization via Information Bottleneck
- 23 January 2026
- Published by Springer Nature
- p. 226-240
DiscoIB: Disentangled Subject Customization via Information Bottleneck
- 12 January 2026
- Published by Springer Nature
- p. 48-62
Beyond Generation: Rethinking Denoising Process for Diffusion Forgery Detection
- 12 January 2026
- Published by Springer Nature
- p. 48-62
Beyond Generation: Rethinking Denoising Process for Diffusion Forgery Detection
- 23 October 2025
- Published by Universidad Internacional de La Rioja in International Journal of Interactive Multimedia and Artificial Intelligence
- Vol. 9 (5) , 78-91
Geometrics Assisted Rubbing Generation and Semantics Enhanced Detection for Small and Dense OBI Character
Character detection is essential for subsequent Oracle Bone Inscription (OBI) research. However, the lack of labeled data and the complexity of small and dense OBI characters are the main difficulties in OBI detection research. In this paper, we propose a framework for rubbing generation that can automatically build up largescale rubbing samples with verisimilar scenarios to noisy wild OBI through geometric and morphological construction combined with style transferring. Moreover, we propose a semantic-enhanced detection model aiming at small and dense OBI through the fusion of multi-resolution feature maps with the enriched feature in the YOLOv5s backbone. We introduce the higher resolution and the Soft-NMS into the proposed OBI detection model to solve the overlapping of small and dense OBI characters. The augmented dataset improves the performance of benchmark object detection models in the real OBI detection task when sufficient data is lacking. Furthermore, the proposed OBI detection model can provide easy and preferable access to OBI detection even with a small number of labeled data and obtain preferable results. Experiments ascertain the effectiveness of the proposed OBI generation framework and the proposed OBI detection model.
- 23 October 2025
- Published by Universidad Internacional de La Rioja in International Journal of Interactive Multimedia and Artificial Intelligence
- Vol. 9 (5) , 78-91
Geometrics Assisted Rubbing Generation and Semantics Enhanced Detection for Small and Dense OBI Character
Character detection is essential for subsequent Oracle Bone Inscription (OBI) research. However, the lack of labeled data and the complexity of small and dense OBI characters are the main difficulties in OBI detection research. In this paper, we propose a framework for rubbing generation that can automatically build up largescale rubbing samples with verisimilar scenarios to noisy wild OBI through geometric and morphological construction combined with style transferring. Moreover, we propose a semantic-enhanced detection model aiming at small and dense OBI through the fusion of multi-resolution feature maps with the enriched feature in the YOLOv5s backbone. We introduce the higher resolution and the Soft-NMS into the proposed OBI detection model to solve the overlapping of small and dense OBI characters. The augmented dataset improves the performance of benchmark object detection models in the real OBI detection task when sufficient data is lacking. Furthermore, the proposed OBI detection model can provide easy and preferable access to OBI detection even with a small number of labeled data and obtain preferable results. Experiments ascertain the effectiveness of the proposed OBI generation framework and the proposed OBI detection model.
- 10 October 2025
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 112
Prior-enhanced attribute embedding for predicting chemoradiotherapy sensitivity in SNSCC patients
- 10 October 2025
- Published by Elsevier in Biomedical Signal Processing and Control
- Vol. 112
Prior-enhanced attribute embedding for predicting chemoradiotherapy sensitivity in SNSCC patients
- 23 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 32 (10709908) , 3027-3031
Mutual Information Disentanglement Based Domain Adaptation Model for EEG Emotion Recognition
To address the lack of generalized feature representation in cross-domain electroencephalogram emotion recognition, this letter proposes a multi-source domain adaptation model that integrates prototype-based class-level constraints into a mutual information disentanglement mechanism. The model mainly consists of a mutual information disentanglement module that separates domain-related and domain-unrelated features by minimizing mutual information, and a prototype classification module that enhances intra-class compactness and semantic consistency across domains. To align multiple source domains with the target domain, Central Moment Discrepancy is minimized in the prototype space. Unlike prior multi-branch MSDA models, our method uses a unified feature extractor, enhancing scalability. Experiments show our model outperforms existing domain adaptation models and achieves high accuracy and stability.
- 23 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Signal Processing Letters
- Vol. 32 (10709908) , 3027-3031
Mutual Information Disentanglement Based Domain Adaptation Model for EEG Emotion Recognition
To address the lack of generalized feature representation in cross-domain electroencephalogram emotion recognition, this letter proposes a multi-source domain adaptation model that integrates prototype-based class-level constraints into a mutual information disentanglement mechanism. The model mainly consists of a mutual information disentanglement module that separates domain-related and domain-unrelated features by minimizing mutual information, and a prototype classification module that enhances intra-class compactness and semantic consistency across domains. To align multiple source domains with the target domain, Central Moment Discrepancy is minimized in the prototype space. Unlike prior multi-branch MSDA models, our method uses a unified feature extractor, enhancing scalability. Experiments show our model outperforms existing domain adaptation models and achieves high accuracy and stability.