English

Prompt-aligned Gradient for Prompt Tuning

Computer Vision and Pattern Recognition 2025-08-14 v4

Abstract

Thanks to the large pre-trained vision-language models (VLMs) like CLIP, we can craft a zero-shot classifier by "prompt", e.g., the confidence score of an image being "[CLASS]" can be obtained by using the VLM provided similarity measure between the image and the prompt sentence "a photo of a [CLASS]". Therefore, prompt shows a great potential for fast adaptation of VLMs to downstream tasks if we fine-tune the prompt-based similarity measure. However, we find a common failure that improper fine-tuning may not only undermine the prompt's inherent prediction for the task-related classes, but also for other classes in the VLM vocabulary. Existing methods still address this problem by using traditional anti-overfitting techniques such as early stopping and data augmentation, which lack a principled solution specific to prompt. We present Prompt-aligned Gradient, dubbed ProGrad, to prevent prompt tuning from forgetting the the general knowledge learned from VLMs. In particular, ProGrad only updates the prompt whose gradient is aligned (or non-conflicting) to the "general direction", which is represented as the gradient of the KL loss of the pre-defined prompt prediction. Extensive experiments demonstrate the stronger few-shot generalization ability of ProGrad over state-of-the-art prompt tuning methods. Codes are available at https://github.com/BeierZhu/Prompt-align.

Keywords

Cite

@article{arxiv.2205.14865,
  title  = {Prompt-aligned Gradient for Prompt Tuning},
  author = {Beier Zhu and Yulei Niu and Yucheng Han and Yue Wu and Hanwang Zhang},
  journal= {arXiv preprint arXiv:2205.14865},
  year   = {2025}
}

Comments

ICCV2023

R2 v1 2026-06-24T11:32:42.088Z