AI RESEARCH

SIGMA: Bridging Structural and Distributional Gaps for Vision Foundation Model Adaptation

arXiv CS.CV

ArXi:2605.27893v1 Announce Type: new Vision Foundation Models (VFMs) have nstrated impressive representational capabilities. However, adapting them to downstream tasks via full fine-tuning incurs prohibitive computational and storage overhead. Parameter-Efficient Fine-Tuning (PEFT) has emerged as a compelling alternative, aiming to achieve performance parity with full fine-tuning at minimal