Research map / Image generation / GANs
A Closer Look at GAN Priors: Exploiting Intermediate Features for Enhanced Model Inversion Attacks
Yixiang Qiu, Hao Fang, Hongyao Yu, Bin Chen, Meikang Qiu, Shu-Tao Xia
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add intermediate feature optimization on top of latent-space search (ablation) | IF-GMI · CelebA ResNet-152, FFHQ prior | Acc@1: 0.803 -> 0.947 | Improved |
| add l1 ball constraint on intermediate features (ablation) | IF-GMI · CelebA ResNet-152, FFHQ prior | FID: 37.528 -> 37.461 | Improved |
| add l1 ball constraint on intermediate features (ablation) | IF-GMI · CelebA ResNet-152, FFHQ prior | Acc@1: 0.945 -> 0.947 | No clear effect |
| optimize three intermediate features instead of latent only (ablation) | IF-GMI · FaceScrub-style ablation, first 100 classes | Acc@1: 0.713 -> 0.846 | Improved |
| choose a later single intermediate block (N=7 vs N=2) (ablation) | IF-GMI (L=1) · first 100 classes | FID: 44.435 -> 57.009 | Hurt |
| use IF-GMI instead of PPA with a MetFaces (OOD) GAN prior | PPA · FaceScrub ResNet-18 | Acc@1: 0.628 -> 0.949 | Improved |
| use IF-GMI instead of PPA with MetFaces prior against CelebA ResNet-152 | PPA · CelebA ResNet-152 | Acc@1: 0.396 -> 0.784 | Improved |
| use IF-GMI instead of PPA on FID with MetFaces prior | PPA · FaceScrub ResNet-18, MetFaces | FID: 62.518 -> 68.107 | Hurt |
| use IF-GMI instead of PPA under BiDO defense | PPA · FaceScrub ResNet-152 + BiDO | Acc@1: 0.356 -> 0.765 | Improved |
| use IF-GMI instead of PPA on dog breed classification | PPA · Stanford Dogs ResNet-152 | Acc@1: 0.950 -> 0.982 | Improved |
Closest papers by meaning
- Knowledge-Enriched Distributional Model Inversion Attacks (ICCV 2021 · similarity 0.92)
- Pseudo Label-Guided Model Inversion Attack via Conditional Generative Adversarial Network (AAAI 2023 · similarity 0.92)
- The Secret Revealer: Generative Model-Inversion Attacks Against Deep Neural Networks (CVPR 2020 · similarity 0.91)
- Pseudo-Private Data Guided Model Inversion Attacks (NeurIPS 2024 · similarity 0.89)
- A Sample-Level Evaluation and Generative Framework for Model Inversion Attacks (AAAI 2025 · similarity 0.87)
- Generative Model Inversion Through the Lens of the Manifold Hypothesis (NeurIPS 2025 · similarity 0.87)
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.
