AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

Contrastive Spectral Rectification: Test-Time Defense towards Zero-shot Adversarial Robustness of CLIP

Sen Nie, Jie Zhang, Zhuo Wang, Shiguang Shan, Xilin CHEN

ICML 2026 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training with feature consistency · 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
add Repulsion loss to Attraction loss (ablation)CSR · General datasetsrobust accuracy (%): 51.6 -> 69.8Improved
add Attraction loss to Repulsion loss (ablation)CSR · General datasetsrobust accuracy (%): 44.7 -> 69.8Improved
add Repulsion loss to Attraction loss (ablation)CSR · Domain datasetsrobust accuracy (%): 23.3 -> 40.3Improved
add Greedy Selection (ablation)CSR · General datasetsrobust accuracy (%): 68.5 -> 69.8Improved
add Greedy Selection (ablation)CSR · Domain datasetsrobust accuracy (%): 39.9 -> 40.3Improved
add Greedy Selection (ablation)CSR · General datasetsclean accuracy (%): 78.6 -> 78.7No clear effect
use CSR test-time defense instead of TTCCLIP ViT-B/16 · 16 datasets avgrobust accuracy (%): 24.5 -> 58.1Improved
use CSR instead of adversarial fine-tuningCLIP ViT-B/16 · 16 datasets avgclean accuracy (%): 43.6 -> 62.9Improved
use CSR defense on segmentationCLIP zero-shot segmentation · VOC2010 under PGDmIoU: 20.8 -> 37.5Improved
use CSR defense on captioningLLaVA · MS-COCO under V-Attackrobust accuracy (%): 21.3 -> 64.3Improved
use CSR defense on VQALLaVA · VQA under V-Attackrobust accuracy (%): 22.6 -> 65.6Improved
use TTC defense on captioning and VQALLaVA · captioning and VQArobust accuracy (%): 21.3 -> 24.1No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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