Chongyang Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Chongyang Li has a total of 32 co-authors. Here are the top co-authors by number of shared publications:
- Wenkai Zhang
Chinese Academy of Sciences, Aerospace Information Research Institute, Beijing, China
6 shared publications - Xian Sun
Chinese Academy of Sciences
6 shared publications - Boyuan Tong
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, China
4 shared publications - Hanbo Bi
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, China
4 shared publications Hongqi Wang
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, China
4 shared publications- Jihao Li
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing, China
3 shared publications
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 12969-12978
RAVA: Retrieval-Augmented Viewpoint Alignment for Subject-Driven Image Generation
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 12969-12978
Open AccessRAVA: Retrieval-Augmented Viewpoint Alignment for Subject-Driven Image Generation
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 79-87
How Much Visual Evidence Does a Grounding Agent Need? Budget Profiles for Agentic Remote-Sensing Visual Grounding
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 79-87
Open AccessHow Much Visual Evidence Does a Grounding Agent Need? Budget Profiles for Agentic Remote-Sensing Visual Grounding
- 3 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-20
ViRefSAM: Visual Reference-Guided Segment Anything Model for Remote Sensing Segmentation
The Segment Anything Model (SAM), with its prompt-driven paradigm, exhibits strong generalization in generic segmentation tasks. However, applying SAM to remote sensing (RS) images still faces two major challenges. First, manually constructing precise prompts for each image (e.g., points or boxes) is labor-intensive and inefficient, especially in RS scenarios with dense small objects or spatially fragmented distributions. Second, SAM lacks domain adaptability, as it is pre-trained primarily on natural images and struggles to capture RS-specific semantics and spatial characteristics, especially when segmenting novel or unseen classes. To address these issues, inspired by few-shot learning, we propose ViRefSAM, a novel framework that guides SAM utilizing only a few annotated reference images that contain class-specific objects. Without requiring manual prompts, ViRefSAM enables automatic segmentation of class-consistent objects across RS images. Specifically, ViRefSAM introduces two key components while keeping SAM’s original architecture intact: (1) a Visual Contextual Prompt Encoder that extracts class-specific semantic clues from reference images and generates object-aware prompts via contextual interaction with target images; and (2) a Dynamic Target Alignment Adapter, integrated into SAM’s image encoder, which mitigates the domain gap by injecting class-specific semantics into target image features, enabling SAM to dynamically focus on task-relevant regions. Extensive experiments on three few-shot segmentation benchmarks, including iSAID-5i, LoveDA-2i, and COCO-20i, demonstrate that ViRefSAM enables accurate and automatic segmentation of unseen classes by leveraging only a few reference images, and consistently outperforms existing few-shot segmentation methods across diverse datasets.
- 3 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-20
ViRefSAM: Visual Reference-Guided Segment Anything Model for Remote Sensing Segmentation
The Segment Anything Model (SAM), with its prompt-driven paradigm, exhibits strong generalization in generic segmentation tasks. However, applying SAM to remote sensing (RS) images still faces two major challenges. First, manually constructing precise prompts for each image (e.g., points or boxes) is labor-intensive and inefficient, especially in RS scenarios with dense small objects or spatially fragmented distributions. Second, SAM lacks domain adaptability, as it is pre-trained primarily on natural images and struggles to capture RS-specific semantics and spatial characteristics, especially when segmenting novel or unseen classes. To address these issues, inspired by few-shot learning, we propose ViRefSAM, a novel framework that guides SAM utilizing only a few annotated reference images that contain class-specific objects. Without requiring manual prompts, ViRefSAM enables automatic segmentation of class-consistent objects across RS images. Specifically, ViRefSAM introduces two key components while keeping SAM’s original architecture intact: (1) a Visual Contextual Prompt Encoder that extracts class-specific semantic clues from reference images and generates object-aware prompts via contextual interaction with target images; and (2) a Dynamic Target Alignment Adapter, integrated into SAM’s image encoder, which mitigates the domain gap by injecting class-specific semantics into target image features, enabling SAM to dynamically focus on task-relevant regions. Extensive experiments on three few-shot segmentation benchmarks, including iSAID-5i, LoveDA-2i, and COCO-20i, demonstrate that ViRefSAM enables accurate and automatic segmentation of unseen classes by leveraging only a few reference images, and consistently outperforms existing few-shot segmentation methods across diverse datasets.
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-17
MCASEN: Boosting Referring Remote Sensing Image Segmentation With Multimodal Semantic Guided Decoding and Recursive Multiscale Fusion
Referring remote sensing image segmentation (RRSIS) aims to segment the target described by textual descriptions in remote sensing (RS) images. Existing methods suffer from insufficient utilization of multimodal contextual information due to the lack of textual semantic guidance during the decoding stage, while their coarse multiscale feature fusion strategies lead to information loss and spatial structure distortion. To mitigate these limitations, we propose the multimodal context-aware semantic enhancement network (MCASEN). MCASEN comprises three key components. First, the metacognitive recursive fusion network (MCRFNet) leverages bidirectional information flow inspired by the human visual cortex to incorporate cross-hierarchical recursive feedback into multiscale feature fusion, allowing high-level semantics to iteratively refine low-level spatial details and thereby mitigating information loss. Second, the spatial perception fusion decoder (SPFD) addresses the absence of textual context during decoding by hierarchically reconstructing multiscale multimodal features and introducing adaptive rotational convolution, which preserves orientation invariance and enhances spatial structure. Finally, the multidimensional guidance fusion module (MGFM) strengthens visual feature representations through multimodal semantic enhancement and the integration of multiscale and multidirectional visual cues. We evaluated the proposed method on two public RRSIS datasets, where our method consistently achieved state-of-the-art (SOTA) performance across multiple metrics compared to existing approaches.
- 28 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-17
MCASEN: Boosting Referring Remote Sensing Image Segmentation With Multimodal Semantic Guided Decoding and Recursive Multiscale Fusion
Referring remote sensing image segmentation (RRSIS) aims to segment the target described by textual descriptions in remote sensing (RS) images. Existing methods suffer from insufficient utilization of multimodal contextual information due to the lack of textual semantic guidance during the decoding stage, while their coarse multiscale feature fusion strategies lead to information loss and spatial structure distortion. To mitigate these limitations, we propose the multimodal context-aware semantic enhancement network (MCASEN). MCASEN comprises three key components. First, the metacognitive recursive fusion network (MCRFNet) leverages bidirectional information flow inspired by the human visual cortex to incorporate cross-hierarchical recursive feedback into multiscale feature fusion, allowing high-level semantics to iteratively refine low-level spatial details and thereby mitigating information loss. Second, the spatial perception fusion decoder (SPFD) addresses the absence of textual context during decoding by hierarchically reconstructing multiscale multimodal features and introducing adaptive rotational convolution, which preserves orientation invariance and enhances spatial structure. Finally, the multidimensional guidance fusion module (MGFM) strengthens visual feature representations through multimodal semantic enhancement and the integration of multiscale and multidirectional visual cues. We evaluated the proposed method on two public RRSIS datasets, where our method consistently achieved state-of-the-art (SOTA) performance across multiple metrics compared to existing approaches.
- 6 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-20
CAMformer: A Single-Stage CNN–Transformer Hybrid for Weakly Supervised Semantic Segmentation in Aerial Imagery
Weakly supervised semantic segmentation (WSSS) with image-level labels has emerged as a cost-effective solution for large-scale remote sensing (RS) imagery, yet CNN-based approaches often suffer from limited receptive fields, incomplete activations, and background confusion in complex multi-class scenes. Transformer-based methods alleviate these issues by modeling long-range dependencies, but many existing designs are multi-stage and inefficient, or fail to adequately address noise and contextual bias in RS data. This work proposes CAMformer, a single-stage hybrid CNN–Transformer framework that jointly captures local details and global semantics for more reliable pseudo-label generation. To further improve robustness, we introduce an inter-class decoupling strategy (ICD) to resolve category conflicts, an uncertainty-weighted optimization (UWO) scheme to suppress noisy activations, and a context intervention sample enhancement (CISE) module to mitigate biased contextual dependencies. Extensive experiments on the challenging iSAID and ISPRS Vaihingen datasets demonstrate that CAMformer achieves state-of-the-art performance with 39.5% and 46.4% mIoU respectively, surpassing recent one-stage methods, while cross-domain validation on PASCAL VOC2012 confirms its generalization to a certain extent. These results highlight the effectiveness of combining CNN–Transformer hybrids with context-aware learning for WSSS in RS applications.
- 6 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 64 (01962892) , 1-20
CAMformer: A Single-Stage CNN–Transformer Hybrid for Weakly Supervised Semantic Segmentation in Aerial Imagery
Weakly supervised semantic segmentation (WSSS) with image-level labels has emerged as a cost-effective solution for large-scale remote sensing (RS) imagery, yet CNN-based approaches often suffer from limited receptive fields, incomplete activations, and background confusion in complex multi-class scenes. Transformer-based methods alleviate these issues by modeling long-range dependencies, but many existing designs are multi-stage and inefficient, or fail to adequately address noise and contextual bias in RS data. This work proposes CAMformer, a single-stage hybrid CNN–Transformer framework that jointly captures local details and global semantics for more reliable pseudo-label generation. To further improve robustness, we introduce an inter-class decoupling strategy (ICD) to resolve category conflicts, an uncertainty-weighted optimization (UWO) scheme to suppress noisy activations, and a context intervention sample enhancement (CISE) module to mitigate biased contextual dependencies. Extensive experiments on the challenging iSAID and ISPRS Vaihingen datasets demonstrate that CAMformer achieves state-of-the-art performance with 39.5% and 46.4% mIoU respectively, surpassing recent one-stage methods, while cross-domain validation on PASCAL VOC2012 confirms its generalization to a certain extent. These results highlight the effectiveness of combining CNN–Transformer hybrids with context-aware learning for WSSS in RS applications.
- Open Access
- 23 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 11908-11925
Efficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning Framework
Fine-tuning pre-trained models for remote sensing tasks often demands substantial computational resources. To reduce memory requirements and training costs, this paper proposes a low-memory fine-tuning framework, called efficient side-tuning (EST), for remote sensing downstream tasks. EST attaches a parallel network to the backbone of the model, and only fine-tunes the parameters of the parallel network during the training phase. The proposed EST Block is the main component of the parallel network, which uses the multi-channel adapter fusion module, gate layer and depthwise convolution to achieve feature selection and enhancement effects. In the evaluation, on six remote sensing datasets including object detection and semantic segmentation, EST achieved SOTA performance results using only less than 40% of the memory expenditure of full finetuning, which is better than all current Parameter Efficient FineTuning (PEFT) methods. Additionally, experiments on backbones of various sizes and classes show that the generalizability of EST is also reliable. EST thus offers a highly efficient and effective approach for efficient transfer learning in remote sensing, unlocking new possibilities for advanced remote sensing applications.
- 23 April 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 11908-11925
Open AccessEfficient Side-Tuning for Remote Sensing: A Low-Memory Fine-Tuning Framework
Fine-tuning pre-trained models for remote sensing tasks often demands substantial computational resources. To reduce memory requirements and training costs, this paper proposes a low-memory fine-tuning framework, called efficient side-tuning (EST), for remote sensing downstream tasks. EST attaches a parallel network to the backbone of the model, and only fine-tunes the parameters of the parallel network during the training phase. The proposed EST Block is the main component of the parallel network, which uses the multi-channel adapter fusion module, gate layer and depthwise convolution to achieve feature selection and enhancement effects. In the evaluation, on six remote sensing datasets including object detection and semantic segmentation, EST achieved SOTA performance results using only less than 40% of the memory expenditure of full finetuning, which is better than all current Parameter Efficient FineTuning (PEFT) methods. Additionally, experiments on backbones of various sizes and classes show that the generalizability of EST is also reliable. EST thus offers a highly efficient and effective approach for efficient transfer learning in remote sensing, unlocking new possibilities for advanced remote sensing applications.
- Open Access
- 8 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 3905-3920
LMF-Net: A Learnable Multimodal Fusion Network for Semantic Segmentation of Remote Sensing Data
Semantic segmentation of remote sensing images has produced a significant effect on many applications, such as land cover and land use, smoke detection, etc.. With the ever-growing remote sensing data, fusing multi-modal data from different sensors is a feasible and effective scheme for semantic segmentation task. Deep learning technology has prominently promoted the development of semantic segmentation. However, the majority of current approaches commonly focus more on feature mixing and construct relatively complex architectures. The further mining for cross-modal features is comparatively insufficient in heterogeneous data fusion. Additionally, complex structures also lead to relatively heavy computation burden. Therefore, in this paper, we propose an end-to-end Learnable Multi-modal Fusion Network (LMF-Net) for remote sensing semantic segmentation. Concretely, we first develop a Multi-Scale Pooling Fusion (MSPF) module by leveraging pooling operator. It provides key-value pairs with multi-modal complementary information in a parameter-free manner and assigns them to selfattention layers of different modal branches. Then, to further harness the cross-modal collaborative embeddings/features, we elaborate two learnable fusion modules, Learnable Embedding Fusion (LEF) and Learnable Feature Fusion (LFF). They are able to dynamically adjust the collaborative relationships of different modal embeddings and features in a learnable approach, respectively. Experiments on two well-established benchmark datasets reveal that our LMF-Net possesses superior segmentation behavior and strong generalization capability. In terms of computation complexity, it achieves competitive performance as well. Ultimately, the contribution of each component involved in LMF-Net is evaluated and discussed in detail.
- 8 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 18 (19391404) , 3905-3920
Open AccessLMF-Net: A Learnable Multimodal Fusion Network for Semantic Segmentation of Remote Sensing Data
Semantic segmentation of remote sensing images has produced a significant effect on many applications, such as land cover and land use, smoke detection, etc.. With the ever-growing remote sensing data, fusing multi-modal data from different sensors is a feasible and effective scheme for semantic segmentation task. Deep learning technology has prominently promoted the development of semantic segmentation. However, the majority of current approaches commonly focus more on feature mixing and construct relatively complex architectures. The further mining for cross-modal features is comparatively insufficient in heterogeneous data fusion. Additionally, complex structures also lead to relatively heavy computation burden. Therefore, in this paper, we propose an end-to-end Learnable Multi-modal Fusion Network (LMF-Net) for remote sensing semantic segmentation. Concretely, we first develop a Multi-Scale Pooling Fusion (MSPF) module by leveraging pooling operator. It provides key-value pairs with multi-modal complementary information in a parameter-free manner and assigns them to selfattention layers of different modal branches. Then, to further harness the cross-modal collaborative embeddings/features, we elaborate two learnable fusion modules, Learnable Embedding Fusion (LEF) and Learnable Feature Fusion (LFF). They are able to dynamically adjust the collaborative relationships of different modal embeddings and features in a learnable approach, respectively. Experiments on two well-established benchmark datasets reveal that our LMF-Net possesses superior segmentation behavior and strong generalization capability. In terms of computation complexity, it achieves competitive performance as well. Ultimately, the contribution of each component involved in LMF-Net is evaluated and discussed in detail.
- 26 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62 (01962892) , 1-14
Injecting Linguistic Into Visual Backbone: Query-Aware Multimodal Fusion Network for Remote Sensing Visual Grounding
The Remote Sensing Visual Grounding (RSVG) task focuses on accurately identifying and localizing specific targets in remote sensing (RS) images using descriptive query expressions. Existing methods independently extract visual and textual features, ignoring early complementary information between image and text. This leads to information loss and misalignment, limiting the model’s ability to distinguish similar targets. To address this challenge, we propose the Query-Aware Multimodal Fusion Network (QAMFN), which introduces an innovative Query-Guided Visual Attention (QGVA) mechanism in the early stages of the visual encoder. This mechanism integrates textual information during the early visual feature extraction process, thereby resolving the issue of missing image-text complementary information. QGVA ensures that the visual backbone accurately focuses on local features highly relevant to the query by injecting textual information into the visual encoding process. Additionally, to enhance the model’s ability to integrate multimodal information and adapt to more complex RS images, we introduce the Text-Semantic Attention-Guided Masking (TAM) module. TAM aggregates multimodal features processed by the backbones and filters out redundant information, producing high-quality fused features. Experiments demonstrate that our approach sets a new record on the DIOR-RSVG dataset, improving accuracy to 81.67% (an absolute increase of 4.98%).
- 26 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Geoscience and Remote Sensing
- Vol. 62 (01962892) , 1-14
Injecting Linguistic Into Visual Backbone: Query-Aware Multimodal Fusion Network for Remote Sensing Visual Grounding
The Remote Sensing Visual Grounding (RSVG) task focuses on accurately identifying and localizing specific targets in remote sensing (RS) images using descriptive query expressions. Existing methods independently extract visual and textual features, ignoring early complementary information between image and text. This leads to information loss and misalignment, limiting the model’s ability to distinguish similar targets. To address this challenge, we propose the Query-Aware Multimodal Fusion Network (QAMFN), which introduces an innovative Query-Guided Visual Attention (QGVA) mechanism in the early stages of the visual encoder. This mechanism integrates textual information during the early visual feature extraction process, thereby resolving the issue of missing image-text complementary information. QGVA ensures that the visual backbone accurately focuses on local features highly relevant to the query by injecting textual information into the visual encoding process. Additionally, to enhance the model’s ability to integrate multimodal information and adapt to more complex RS images, we introduce the Text-Semantic Attention-Guided Masking (TAM) module. TAM aggregates multimodal features processed by the backbones and filters out redundant information, producing high-quality fused features. Experiments demonstrate that our approach sets a new record on the DIOR-RSVG dataset, improving accuracy to 81.67% (an absolute increase of 4.98%).