BindCLIP: One Balanced Coupling For Compositional Vision Language Scoring
为了解决视觉-语言模型偏好组合不正确描述的问题，BindCLIP通过一个平衡的token-patch-depth最优传输耦合方法来评分。
0 citationsRead paper
Selected work
Representative Papers
OGR-MARL: Option-Guided Residual Multi-Agent Reinforcement Learning for Heterogeneous USV Cooperative Pursuit in Constrained Port Waterways
This work addresses the challenge of cooperative pursuit by heterogeneous unmanned surface vehicles (USVs) in constrained port waterways, where navigation, traffic, and role-based constraints must be respected. To this end, the authors propose the OGR-MARL framework, which integrates shared goal beliefs, a role-conditioned options mechanism, adaptive rule-based penalties, and residual policy learning. This enables multi-agent systems to refine actions while adhering to prior domain rules, rather than exploring from scratch. Notably, the framework decouples the options mechanism and residual learning from specific multi-agent reinforcement learning algorithms—such as MADDPG, MATD3, MAPPO, and MASAC—yielding a general architecture capable of zero-shot transfer to real-world port maps. Evaluated in the simulated Shin-no-Mon port environment, OGR-MASAC achieves a 75.0% capture rate, substantially outperforming baselines, and successfully generalizes without retraining to realistic scenarios derived from QGIS/AIS data, demonstrating both effective coordination and strict rule compliance.
0 citationsRead paper
BPC-Net: Annotation-Free Skin Lesion Segmentation via Boundary Probability Calibration
This work addresses the challenges of incomplete lesion boundaries in unsupervised skin lesion segmentation, which stem from underconfident boundary probabilities, noisy pseudo-labels, and unstable cross-domain transfer. To tackle these issues, the authors propose BPC-Net, a novel framework that explicitly models and calibrates boundary probabilities through Gaussian probability smoothing to achieve precise boundary recovery. Additionally, BPC-Net incorporates a feature-disentangled decoder and an adaptive transfer strategy that updates only the interaction branch, enhancing segmentation accuracy and deployment stability under fully unsupervised conditions. Evaluated on ISIC-2017, ISIC-2018, and PH2 datasets, BPC-Net achieves state-of-the-art performance among unsupervised methods, with macro-averaged Dice coefficient and Jaccard index reaching 85.80% and 76.97%, respectively—approaching the performance of supervised methods on PH2.
0 citationsRead paper
Caption Generation for Dongba Paintings via Prompt Learning and Semantic Fusion
This work addresses the challenge of domain shift that hinders mainstream image captioning models from accurately capturing the nuanced cultural semantics embedded in Dongba paintings. To this end, the authors propose the PVGF-DPC framework, which integrates a culture-aware content prompting mechanism to steer caption generation and introduces a vision-semantic–generation fusion loss to align visual features with culturally specific themes. The model leverages MobileNetV2 for visual feature extraction and a 10-layer Transformer decoder initialized with BERT, enhanced by prompt-guided decoding and joint optimization. Evaluated on a newly constructed dataset of 9,408 augmented Dongba images, the approach demonstrates significant improvements in both semantic accuracy and cultural relevance across seven Dongba thematic categories.
0 citationsRead paper
Multi-View Deformable Convolution Meets Visual Mamba for Coronary Artery Segmentation
This work addresses the challenges of coronary artery segmentation in CT angiography, where the vessels exhibit highly elongated and complex branching structures alongside extreme foreground-background imbalance. Conventional CNNs struggle to capture long-range dependencies, while Vision Transformers incur prohibitive computational costs for clinical deployment. To overcome these limitations, we propose MDSVM-UNet, a two-stage segmentation framework: in the encoding stage, a multi-directional snake-like deformable convolution (MDSConv) fuses features from three orthogonal planes; in the decoding stage, a residual Visual Mamba (RVM) module models inter-slice long-range dependencies with linear complexity. Coupled with a coarse-to-fine patching strategy, our method effectively recovers fine vessels while suppressing false positives. This is the first approach to integrate multi-view deformable convolution with Visual Mamba, achieving significantly improved segmentation accuracy under low computational overhead—making it suitable for resource-constrained clinical settings.
0 citationsRead paper
Recent publications
Latest Papers
BindCLIP: One Balanced Coupling For Compositional Vision Language Scoring
为了解决视觉-语言模型偏好组合不正确描述的问题，BindCLIP通过一个平衡的token-patch-depth最优传输耦合方法来评分。
0 citationsRead paper
OGR-MARL: Option-Guided Residual Multi-Agent Reinforcement Learning for Heterogeneous USV Cooperative Pursuit in Constrained Port Waterways
This work addresses the challenge of cooperative pursuit by heterogeneous unmanned surface vehicles (USVs) in constrained port waterways, where navigation, traffic, and role-based constraints must be respected. To this end, the authors propose the OGR-MARL framework, which integrates shared goal beliefs, a role-conditioned options mechanism, adaptive rule-based penalties, and residual policy learning. This enables multi-agent systems to refine actions while adhering to prior domain rules, rather than exploring from scratch. Notably, the framework decouples the options mechanism and residual learning from specific multi-agent reinforcement learning algorithms—such as MADDPG, MATD3, MAPPO, and MASAC—yielding a general architecture capable of zero-shot transfer to real-world port maps. Evaluated in the simulated Shin-no-Mon port environment, OGR-MASAC achieves a 75.0% capture rate, substantially outperforming baselines, and successfully generalizes without retraining to realistic scenarios derived from QGIS/AIS data, demonstrating both effective coordination and strict rule compliance.
0 citationsRead paper
BPC-Net: Annotation-Free Skin Lesion Segmentation via Boundary Probability Calibration
This work addresses the challenges of incomplete lesion boundaries in unsupervised skin lesion segmentation, which stem from underconfident boundary probabilities, noisy pseudo-labels, and unstable cross-domain transfer. To tackle these issues, the authors propose BPC-Net, a novel framework that explicitly models and calibrates boundary probabilities through Gaussian probability smoothing to achieve precise boundary recovery. Additionally, BPC-Net incorporates a feature-disentangled decoder and an adaptive transfer strategy that updates only the interaction branch, enhancing segmentation accuracy and deployment stability under fully unsupervised conditions. Evaluated on ISIC-2017, ISIC-2018, and PH2 datasets, BPC-Net achieves state-of-the-art performance among unsupervised methods, with macro-averaged Dice coefficient and Jaccard index reaching 85.80% and 76.97%, respectively—approaching the performance of supervised methods on PH2.
0 citationsRead paper
Caption Generation for Dongba Paintings via Prompt Learning and Semantic Fusion
This work addresses the challenge of domain shift that hinders mainstream image captioning models from accurately capturing the nuanced cultural semantics embedded in Dongba paintings. To this end, the authors propose the PVGF-DPC framework, which integrates a culture-aware content prompting mechanism to steer caption generation and introduces a vision-semantic–generation fusion loss to align visual features with culturally specific themes. The model leverages MobileNetV2 for visual feature extraction and a 10-layer Transformer decoder initialized with BERT, enhanced by prompt-guided decoding and joint optimization. Evaluated on a newly constructed dataset of 9,408 augmented Dongba images, the approach demonstrates significant improvements in both semantic accuracy and cultural relevance across seven Dongba thematic categories.
0 citationsRead paper
Multi-View Deformable Convolution Meets Visual Mamba for Coronary Artery Segmentation
This work addresses the challenges of coronary artery segmentation in CT angiography, where the vessels exhibit highly elongated and complex branching structures alongside extreme foreground-background imbalance. Conventional CNNs struggle to capture long-range dependencies, while Vision Transformers incur prohibitive computational costs for clinical deployment. To overcome these limitations, we propose MDSVM-UNet, a two-stage segmentation framework: in the encoding stage, a multi-directional snake-like deformable convolution (MDSConv) fuses features from three orthogonal planes; in the decoding stage, a residual Visual Mamba (RVM) module models inter-slice long-range dependencies with linear complexity. Coupled with a coarse-to-fine patching strategy, our method effectively recovers fine vessels while suppressing false positives. This is the first approach to integrate multi-view deformable convolution with Visual Mamba, achieving significantly improved segmentation accuracy under low computational overhead—making it suitable for resource-constrained clinical settings.
0 citationsRead paper