A unified Diffusion Transformer framework for semantic and controllable image fusion, supporting multiple fusion tasks and extending image fusion toward text-controllable fusion and multimodal segmentation.
A referring-image-segmentation perspective on text-driven infrared-visible image fusion, improving interpretability and downstream consistency.
A coupled image fusion and rain-removal framework that improves robustness for infrared-visible perception in rainy scenes.
A coupled fusion and dehazing framework for robust visible-infrared image fusion in hazy environments.
A foveated vision approach for ultra-high-resolution remote sensing reasoning, dynamically selecting key regions for fine-grained attention.
A training-free pipeline for ultra-high-resolution remote sensing VQA that adaptively zooms into key regions, reducing token and memory cost while improving reasoning efficiency.
A structured policy learning framework for geospatial pixel reasoning, improving language-to-pixel segmentation generalization with reduced reliance on dense mask supervision.
A multimodal embodied reasoning benchmark for evaluating models across exploration, dynamic spatial-semantic reasoning, and multi-step task execution.
