CoDA: Co-adaptive Dual-path Alignment for Vision-Language Models

Yi Zhang; Rui Zhu; Channi Li; Xiaoxu Li; Zhanyu Ma; Jing-Hao Xue

IJCAI 20260 citations

CoDA: Co-adaptive Dual-path Alignment for Vision-Language Models

Yi Zhang, Rui Zhu, Channi Li, Xiaoxu Li, Zhanyu Ma, Jing-Hao Xue

Abstract

Parameter-efficient adaptation methods, such as adapters and prompt learning, have become popular for transferring pre-trained vision-language models (VLMs) to downstream tasks. However, under limited supervision during adaptation, overly aggressive cross-modal alignment can distort the intrinsic structure of modality-specific representations, leading to degraded generalization. In this paper, we propose Co-adaptive Dual-path Alignment (CoDA), a new adaptation framework that explicitly disentangles and coordinates cross-modal semantic alignment and intra-modal structural consistency. In CoDA, we also propose a parent class labeling strategy that injects hierarchical semantic priors into textual prompts, further stabilizing alignment under limited supervision. Extensive experiments on eleven benchmark datasets show that CoDA outperforms state-of-the-art parameter-efficient methods, particularly under few-shot learning and distribution-shift scenarios. Our code is available at https://github.com/yizhang-ac/CoDA.

Computer Vision: Multimodal learningComputer Vision: Transfer, low-shot, semi- and un- supervised learningMachine Learning: Few-shot learning

BibTeX

@inproceedings{ijcai2026_codacoadaptivedu,
  title = {CoDA: Co-adaptive Dual-path Alignment for Vision-Language Models},
  author = {Yi Zhang and Rui Zhu and Channi Li and Xiaoxu Li and Zhanyu Ma and Jing-Hao Xue},
  booktitle = {IJCAI 2026},
  year = {2026}
}