SARVLM: Vision-Language Model for SAR Imagery Understanding

Researchers have developed SARVLM, a vision-language foundation model for semantic understanding of Synthetic Aperture Radar (SAR) imagery. SAR is valued for its all-weather imaging capability, but existing SAR foundation models focus on low-level visual features and neglect multi-modal representation. To address this, the team constructed SARVLM-1M, a large-scale dataset of over one million image-text pairs aggregated from existing sources. They also proposed a two-stage domain transfer training strategy using optical remote sensing data as an intermediate bridge to transfer knowledge from natural images to SAR domains. The work is detailed in a paper on arXiv (2510.22665).