AI Paper Atlas

Research map / Federated learning and privacy / Differential privacy

COVERT: Privacy-Preserving Covariant Obfuscation for VLMaaS via Exact Reparameterization and Tailored Tuning

Wenqiang Ruan, ZiRui Huang, Yu Lin, Qizhi Zhang, Yunlong Mao, Quanwei Cai, Jue Hong, Sheng Zhong, Ye Wu

ECCV 2026 · Federated learning and privacy › Differential privacy · approach: Differentially private synthetic data · 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 tailored utility-aware tuning of first ViT block attention (ablation)COVERT obfuscation · ViT hidden-state inversion (adaptive attack), layer 1SSIM: 0.3404 -> 0.1550Improved
add tailored utility-aware tuning of first ViT block attention (ablation)COVERT obfuscation · ViT hidden-state inversion (adaptive attack), layer 1MSE: 3275.28 -> 4577.28Improved
scale model from 7B to 32BCOVERT · Five VLM benchmarks (avg of 5 datasets)average accuracy drop: 2.58 -> 1.83Improved
scale model from 7B to 72BCOVERT · Five VLM benchmarks (avg of 5 datasets)average accuracy drop: 2.58 -> 2.28Improved
add 8-bit quantization of obfuscated modelCOVERT (Qwen2.5-VL-7B) · Five VLM benchmarks (avg of 5 datasets)average accuracy: 68.63 -> 68.40No clear effect
apply COVERT to Fara-7B and Mimo-7B variantsCOVERT · Five VLM benchmarks (avg of 5 datasets)average accuracy drop: 1.53 and 2.81 dropNo clear effect
replace DP-Forward (eps=128) with COVERTDP-Forward · Dataset 1 (Qwen2.5-VL-7B)accuracy: 0.07 -> 79.79Improved
replace DP-Forward (eps=128) with COVERTDP-Forward · Dataset 4 (Qwen2.5-VL-7B)accuracy: 13.76 -> 83.87Improved
add COVERT obfuscation to serving Qwen2.5-VL-32B on vLLMPlaintext inference · Online inference, Qwen2.5-VL-32B vLLMthroughput: 25.17 -> 23.76Hurt
add COVERT obfuscation to serving Qwen2.5-VL-32B on vLLMPlaintext inference · Online inference, Qwen2.5-VL-32B vLLMTPOT: 26.73 -> 29.26Hurt
add 64-bin quantization and minimal patch expansion (ablation)COVERT communication · Communication expansionexpansion ratio: 7.68x -> 1.30xImproved
send obfuscated image as JPG instead of PNG (ablation)COVERT communication · Communication expansionexpansion ratio: 7.68x -> 32.15xHurt

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