Xiaodong Gu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Xiaodong Gu has a total of 71 co-authors. Here are the top co-authors by number of shared publications:
Yuanyuan Wang
Department of Electronic Engineering, Fudan University, Shanghai 200433, China
9 shared publications- Qingrong Cheng
Department of Electronic Engineering, Fudan University, Shanghai, China
7 shared publications - Haibing Wu
Department of Electronic Engineering Fudan University 200433 Shanghai China
6 shared publications Mengmeng Liao
Department of Electronic Engineering Fudan University Shanghai 200433 People's Republic of China
6 shared publications- Qi Guo
Fudan University
6 shared publications - Zhenshan Tan
School of Computer Science, Nanjing University of Information Science and Technology, Nanjing, China
6 shared publications
Latest Publications
- 2 May 2026
- Published by Springer Nature in Applied Intelligence
- Vol. 56 (7)
Audio–visual segmentation via hierarchical side tuning with state space model
- 2 May 2026
- Published by Springer Nature in Applied Intelligence
- Vol. 56 (7)
Audio–visual segmentation via hierarchical side tuning with state space model
- 2 February 2026
- Published by Elsevier in Neurocomputing
- Vol. 674
Vision-language alignment with sigmoid loss and dual-token contrastive change localizer for precise change captioning
- 2 February 2026
- Published by Elsevier in Neurocomputing
- Vol. 674
Vision-language alignment with sigmoid loss and dual-token contrastive change localizer for precise change captioning
- 3 January 2026
- Published by Elsevier in Big Data Research
- Vol. 43
Efficient time series forecasting with gated attention and patched data: A transformer-based approach
- 3 January 2026
- Published by Elsevier in Big Data Research
- Vol. 43
Efficient time series forecasting with gated attention and patched data: A transformer-based approach
- 24 December 2025
- Published by Springer Nature in International Journal of Computer Vision
- Vol. 134 (1) , 11
MMRL++: Parameter-Efficient and Interaction-Aware Representation Learning for Vision-Language Models
- 24 December 2025
- Published by Springer Nature in International Journal of Computer Vision
- Vol. 134 (1) , 11
MMRL++: Parameter-Efficient and Interaction-Aware Representation Learning for Vision-Language Models
- 19 December 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 200, 102-107
Enhanced facial expression manipulation through domain-aware transformation and dual-level classification with expression awarness loss in the CLIP space
- 19 December 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 200, 102-107
Enhanced facial expression manipulation through domain-aware transformation and dual-level classification with expression awarness loss in the CLIP space
- 23 November 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 301
Knowledge distillation from single-label to multi-label with class activation maps
- 23 November 2025
- Published by Elsevier in Expert Systems with Applications
- Vol. 301
Knowledge distillation from single-label to multi-label with class activation maps
- Open Access
MMRL++: Parameter-Efficient and Interaction-Aware Representation Learning for Vision-Language Models
Large-scale pre-trained Vision-Language Models (VLMs) have significantly advanced transfer learning across diverse tasks. However, adapting these models with limited few-shot data often leads to overfitting, undermining their ability to generalize to new tasks. To address this, we propose Multi-Modal Representation Learning (MMRL), which introduces a shared, learnable, modality-agnostic representation space. MMRL generates space tokens projected into both text and image encoders as representation tokens, enabling more effective cross-modal interactions. Unlike prior methods that mainly optimize class token features, MMRL inserts representation tokens into higher encoder layers--where task-specific features are more prominent--while preserving general knowledge in the lower layers. During training, both class and representation features are jointly optimized: a trainable projection layer is applied to representation tokens for task adaptation, while the projection layer for class token remains frozen to retain pre-trained knowledge. To further promote generalization, we introduce a regularization term aligning class and text features with the frozen VLM's zero-shot features. At inference, a decoupling strategy uses both class and representation features for base tasks, but only class features for novel tasks due to their stronger generalization. Building upon this, we propose MMRL++, a parameter-efficient and interaction-aware extension that significantly reduces trainable parameters and enhances intra-modal interactions--particularly across the layers of representation tokens--allowing gradient sharing and instance-specific information to propagate more effectively through the network. Extensive experiments on 15 datasets demonstrate that MMRL and MMRL++ consistently outperform state-of-the-art methods, achieving a strong balance between task-specific adaptation and generalization.
Open AccessMMRL++: Parameter-Efficient and Interaction-Aware Representation Learning for Vision-Language Models
Large-scale pre-trained Vision-Language Models (VLMs) have significantly advanced transfer learning across diverse tasks. However, adapting these models with limited few-shot data often leads to overfitting, undermining their ability to generalize to new tasks. To address this, we propose Multi-Modal Representation Learning (MMRL), which introduces a shared, learnable, modality-agnostic representation space. MMRL generates space tokens projected into both text and image encoders as representation tokens, enabling more effective cross-modal interactions. Unlike prior methods that mainly optimize class token features, MMRL inserts representation tokens into higher encoder layers--where task-specific features are more prominent--while preserving general knowledge in the lower layers. During training, both class and representation features are jointly optimized: a trainable projection layer is applied to representation tokens for task adaptation, while the projection layer for class token remains frozen to retain pre-trained knowledge. To further promote generalization, we introduce a regularization term aligning class and text features with the frozen VLM's zero-shot features. At inference, a decoupling strategy uses both class and representation features for base tasks, but only class features for novel tasks due to their stronger generalization. Building upon this, we propose MMRL++, a parameter-efficient and interaction-aware extension that significantly reduces trainable parameters and enhances intra-modal interactions--particularly across the layers of representation tokens--allowing gradient sharing and instance-specific information to propagate more effectively through the network. Extensive experiments on 15 datasets demonstrate that MMRL and MMRL++ consistently outperform state-of-the-art methods, achieving a strong balance between task-specific adaptation and generalization.
- 3 April 2025
- Published by Springer Nature in Applied Intelligence
- Vol. 55 (7) , 1-14
Intra-frame scan-free video state spaces model for video moment retrieval
- 3 April 2025
- Published by Springer Nature in Applied Intelligence
- Vol. 55 (7) , 1-14
Intra-frame scan-free video state spaces model for video moment retrieval
- 8 December 2024
- Published by Springer Nature
- p. 391-408
A StyleCLIP-Based Facial Emotion Manipulation Method for Discrepant Emotion Transitions
- 8 December 2024
- Published by Springer Nature
- p. 391-408
A StyleCLIP-Based Facial Emotion Manipulation Method for Discrepant Emotion Transitions
- 5 October 2024
- Published by Elsevier in Information Systems
- Vol. 128
Exploiting explicit item–item correlations from knowledge graphs for enhanced sequential recommendation
- 5 October 2024
- Published by Elsevier in Information Systems
- Vol. 128
Exploiting explicit item–item correlations from knowledge graphs for enhanced sequential recommendation