Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Finetune Like You Pretrain: Boosting Zero-shot Adversarial Robustness in Vision-language Models

About

Despite their impressive zero-shot abilities, vision-language models such as CLIP have been shown to be susceptible to adversarial attacks. To enhance its adversarial robustness, recent studies finetune the pretrained vision encoder of CLIP with adversarial examples on a proxy dataset such as ImageNet by aligning adversarial images with correct class labels. However, these methods overlook the important roles of training data distributions and learning objectives, resulting in reduced zero-shot capabilities and limited transferability of robustness across domains and datasets. In this work, we propose a simple yet effective paradigm AdvFLYP, which follows the training recipe of CLIP's pretraining process when performing adversarial finetuning to the model. Specifically, AdvFLYP finetunes CLIP with adversarial images created based on image-text pairs collected from the web, and match them with their corresponding texts via a contrastive loss. To alleviate distortion of adversarial image embeddings of noisy web images, we further propose to regularise AdvFLYP by penalising deviation of adversarial image features. We show that logit- and feature-level regularisation terms benefit robustness and clean accuracy, respectively. Extensive experiments on 14 downstream datasets spanning various domains show the superiority of our paradigm over mainstream practices. Our code and model weights are released at https://github.com/Sxing2/AdvFLYP.

Songlong Xing, Weijie Wang, Zhengyu Zhao, Jindong Gu, Philip Torr, Nicu Sebe• 2026

Related benchmarks

TaskDatasetResultRank
Image ClassificationSUN397
Accuracy57.26
450
Image ClassificationStanfordCars
Accuracy25.76
384
Image ClassificationCIFAR100
Accuracy2.44
347
Image ClassificationCIFAR100
Accuracy48.86
301
Image ClassificationOxfordPets
Accuracy83.89
298
Image ClassificationFGVCAircraft
Accuracy16.8
289
Image ClassificationCIFAR10
Accuracy (%)75.76
282
Image ClassificationFood101
Accuracy69.48
177
Image ClassificationCIFAR10
Accuracy52.24
143
Image ClassificationStanford Cars
Top-1 Accuracy48.1
104
Showing 10 of 18 rows

Other info

Follow for update