Ta-Adapter: Enhancing few-shot CLIP with task-aware encoders

Wenbo Zhang, Yifan Zhang, Yuyang Deng, Wenlong Zhang, Jianfeng Lin, Binqiang Huang, Jinlu Zhang, Wenhao Yu

Abstract: Highlights•Our model design combines both the advantages of prompt learning and adapter tuning.•Align CLIP visual and textual encoders with specific datasets via few-shot images.•Our model further enhances CLIP’s few-shot capability, obtaining superior results.