MF-RSVLM: A VLM for Remote Sensing

This paper introduces MF-RSVLM, a vision-language model specifically designed for remote sensing applications. The core contribution lies in its multi-feature fusion approach, which aims to overcome the limitations of existing VLMs in this domain by better capturing fine-grained visual features and mitigating visual forgetting. The model's performance is validated across various remote sensing tasks, demonstrating state-of-the-art or competitive results.

• •Addresses limitations of existing VLMs in remote sensing.
• •Employs a multi-feature fusion approach for better visual feature extraction.
• •Includes a recurrent visual feature injection scheme to reduce visual forgetting.
• •Achieves strong performance on various remote sensing benchmarks.