CVPR Findings2026
Revisiting Model Inversion Evaluation: From Misleading Standards to Reliable Privacy Assessment
Sy-Tuyen Ho, Koh Jun Hao, Ngoc-Bao Nguyen, Alexander Binder, Ngai-Man Cheung
Abstract
Model Inversion (MI) attacks aim to reconstruct information from private training data by exploiting access to a target model. Nearly all recent MI studies evaluate attack success using a standard framework that computes attack accuracy through a secondary evaluation model trained on the same private data and task design as the target model. In this paper, we present the first in-depth analysis of this dominant evaluation framework and reveal a fundamental issue: many reconstructions deemed "successful" under the existing framework are in fact false positives that do not capture the visual identity of the target individual.
As our first contribution, we show that these MI false positives satisfy the same formal conditions as Type I adversarial examples. In addition, through controlled experiments, we demonstrate extremely high false-positive transferability, an empirical signature characteristic of adversarial behavior, indicating that many MI false positives likely contain Type I adversarial features. This adversarial transferability significantly inflates reported attack accuracy and leads to an overstatement of privacy leakage in existing MI work.
To address this issue, as our second contribution, we introduce a new evaluation framework based on Multimodal Large Language Models (MLLMs), whose general-purpose visual reasoning avoids the shared-task vulnerability and reduces Type-I adversarial transferability of current evaluation framework. We propose systematic design principles for MLLM-based evaluation. Using this framework, we reassess 27 MI attack setups across diverse datasets, target models, and priors, and find consistently high false-positive rates under the conventional approach. Our results call for a reevaluation of progress in MI research and establish MLLM-based evaluation as a more reliable standard for assessing privacy risks in machine learning systems.
To support future work, we release our MLLM-based
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
1 top-tier paper has cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- Do Vision-Language Models Leak What They Learn? Adaptive Token-Weighted Model Inversion Attacks cites Neural Network Inversion in Adversarial Setting via Background Knowledge Alignment
- Do Vision-Language Models Leak What They Learn? Adaptive Token-Weighted Model Inversion Attacks cites Variational Model Inversion Attacks
- Do Vision-Language Models Leak What They Learn? Adaptive Token-Weighted Model Inversion Attacks cites Knowledge-Enriched Distributional Model Inversion Attacks
- Neural Network Inversion in Adversarial Setting via Background Knowledge Alignment cites Towards Evaluating the Robustness of Neural Networks
Cited by top-tier papers1
Builds on22
- Towards Evaluating the Robustness of Neural NetworksNicholas Carlini, David A. WagnerS&P 2017 · 9,786 citations
- Mitigating Neural Network Overconfidence with Logit NormalizationHongxin Wei, Renchunzi Xie, Hao Cheng, Lei Feng et al.ICML 2022 · 386 citations
- Neural Network Inversion in Adversarial Setting via Background Knowledge AlignmentZiqi Yang, Jiyi Zhang, Ee-Chien Chang, Zhenkai LiangCCS 2019 · 257 citations
- Variational Model Inversion AttacksKuan-Chieh Wang, Yan Fu, Ke Li, Ashish Khisti et al.NeurIPS 2021 · 142 citations
- Knowledge-Enriched Distributional Model Inversion AttacksSi Chen, Mostafa Kahla, Ruoxi Jia, Guo-Jun QiICCV 2021 · 124 citations
Related papers
- A Sample-Level Evaluation and Generative Framework for Model Inversion AttacksHaoyang Li, Li Bai, Qingqing Ye, Haibo Hu et al.AAAI 2025 · 4 citations
- Model Inversion Robustness: Can Transfer Learning Help?Sy-Tuyen Ho, Koh Jun Hao, Keshigeyan Chandrasegaran, Ngoc-Bao Nguyen et al.CVPR 2024
- The Secret Revealer: Generative Model-Inversion Attacks Against Deep Neural NetworksYuheng Zhang, Ruoxi Jia, Hengzhi Pei, Wenxiao Wang et al.CVPR 2020
- MultiPriv: Benchmarking Individual-Level Privacy Reasoning in Vision-Language ModelsXiongtao Sun, HUI LI, Jiaming Zhang, Yujie Yang et al.ICML 2026 · 3 citations
- Re-Thinking Model Inversion Attacks Against Deep Neural NetworksNgoc-Bao Nguyen, Keshigeyan Chandrasegaran, Milad Abdollahzadeh, Ngai-Man CheungCVPR 2023
