Grounding Before Reporting: Local-to-Global Decoupled Learning for Medical Vision Language Models
Local-to-global decoupled learning for medical report generation, improving both global reporting quality and local anatomical grounding.
Local-to-global decoupled learning for medical report generation, improving both global reporting quality and local anatomical grounding.
High-resolution defect segmentation under severe class imbalance with weighted sampling, composite losses, and sliding-window inference.
Linear-complexity spatiotemporal modeling with semantic modulation for high-resolution remote sensing.
Knowledge-enhanced air-quality forecasting with knowledge graphs, RGCN, and Mamba.
Dual-encoder fine-grained sentiment analysis with dynamic feature fusion.