Layerlink: Bridging remote sensing object detection and large vision models with efficient fine-tuning

Xingkui Zhu, Dingkang Liang, Xingyu Jiang, Yiran Guan, Yuliang Liu, Yingying Zhu, Xiang Bai

Abstract: Highlights•Proposes Layerlink, a PEFT framework for efficient RSOD fine-tuning.•Introduces Conductor Adapter for efficient adaptation with fewer parameters.•Shares adapter modules across LVM layers, further reducing redundancy.•Benchmarks NLP-based PEFT methods comprehensively in RSOD.•Achieves state-of-the-art on DOTA and DIOR-R with less than 8% of parameters tuned.