Filling Before Advancing: Capability-Gap-Driven Post-Training for Scenario-Specialized Remote Sensing MLLMs
Remote sensing multimodal large language models struggle with specific scenario understanding due to limited high-quality data. The authors propose Filling Before Advancing (FBA), a post-training method that sequentially closes capability gaps before specializing in a target scenario. They use a three-stage process—semantic anchoring, domain bridging, and evidence-grounded tuning—and evaluate it on harbor understanding tasks, showing superior performance over direct fine-tuning methods.