Sojung An

LoCA: Spatially-Aware Low-Rank Convolutional Adaptation of Vision Foundation Models

Sojung An, Junha Lee, Sujeong You, Nam Ik Cho, Donghyun Kim

Pre-trained Vision Foundation Models (VFMs) provide strong visual representations for diverse downstream tasks. The key challenge of VFM adaptation stems from the prohibitive costs of full fine-tuning and catastrophic forgetting. To address this, Low-Rank Adaptation (LoRA) has em…Read more