AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

AGFT: Alignment-Guided Fine-Tuning for Zero-Shot Adversarial Robustness of Vision-Language Models

Yubo Cui, Xianchao Guan, Zijun Xiong, Zheng Zhang

CVPR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: Vision-language model robustness · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace hard-label supervision with calibrated soft pre-trained predictions (AGFT) vs TeCoACLIP ViT-B/32 adversarial fine-tuning · 15 zero-shot datasets, PGD-20 eps=1/255average robust accuracy (%): 38.51 -> 46.57Improved
replace hard-label supervision with AGFTCLIP ViT-B/32 adversarial fine-tuning · 15 zero-shot datasets, cleanaverage clean accuracy (%): 56.93 -> 61.35Improved
use AGFT vs strongest baseline GLADIATORCLIP ViT-B/32 · 15 zero-shot datasets, PGD-20 eps=1/255average robust accuracy (%): 43.46 -> 46.57Improved
evaluate at larger inference budget 4/255 (fine-tuned at 1/255)AGFT vs TGA-ZSR · PGD-20 eps=4/255average robust accuracy (%): AGFT 8.88 vs TGA-ZSR 10.30Hurt
fine-tune with larger budget eps=4/255 using AGFTCLIP ViT-B/32 · 15 datasetsaverage clean accuracy (%): best baseline 41.73 -> 49.47Improved
use AGFT on RN50x4 backboneCLIP RN50x4 · 15 datasetsaverage clean accuracy (%): clean 52.04 (TeCoA) vs 51.23; PGD robust 31.00 vs 34.69Mixed
use AGFT on ImageNet in-domainCLIP ViT-B/32 · ImageNet PGD-20robust accuracy (%): TGA-ZSR 52.80 vs AGFT 44.95Hurt
use AGFT on distribution-shifted ImageNet-RCLIP ViT-B/32 · ImageNet-R PGD-20robust accuracy (%): best baseline 41.08 -> 45.21Improved
decrease temperature scaling ratio gamma from 1.0 to 0.4 (calibration) (ablation)AGFT · hyperparameter grid, 1/tau=180clean accuracy (%): 58.96 -> 61.35Improved
decrease gamma from 1.0 to 0.2 at 1/tau=100 (ablation)AGFT · hyperparameter gridrobust accuracy (%): 45.07 -> 33.29Hurt
increase 1/tau from 100 to 220 at gamma=0.4 (ablation)AGFT · hyperparameter gridrobust accuracy (%): robust 43.17 -> 47.08; clean 62.20 -> 60.68Mixed
add one extra forward pass for pre-trained distribution (vs PMG-AFT/TGA-ZSR)AGFT training cost · training time per epochminutes: TGA-ZSR 39.78 -> AGFT 25.96 (TeCoA 23.45)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy