Research map / Image generation / GANs
What can Discriminator do? Towards Box-free Ownership Verification of Generative Adversarial Networks
Ziheng Huang, Boheng Li, Yan Cai, Run Wang, Shangwei Guo, Liming Fang, Jing Chen, Lina Wang
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 |
|---|---|---|---|
| initialize hypersphere network and center from the paired discriminator instead of randomly (ablation) | box-free GAN ownership verification · Ambiguity attack (different models) | AUC (%): 57.21 -> 75.63 | Improved |
| use SOTA GAN architectures (StyleGAN family) rather than simple DCGAN | verification scheme · Different-model AUC | AUC (%): AUC > 80% | Improved |
| apply output image transformations (noise, blur, JPEG, crop) within acceptable quality | verification scheme · CelebA/LSUN | AUC (%): 54.15 to 57.24 AUC across transformations | No clear effect |
| prune GAN weights up to about 40% | verification scheme · DCGAN CelebA 128 | AUC (%): AUC below 60% at about 40% pruning | No clear effect |
| train two GANs on different datasets (disjoint subsets) | verification scheme · CelebA/LSUN | AUC (%): AUC above 75% | Improved |
| models differ only in initialization seed | verification scheme · image synthesis | AUC (%): about 60% vs less than 53% for identical models | Mixed |
Closest papers by meaning
- Rob-GAN: Generator, Discriminator, and Adversarial Attacker (CVPR 2019 · similarity 0.87)
- Mixed batches and symmetric discriminators for GAN training (ICML 2018 · similarity 0.87)
- DuelGAN: A Duel between Two Discriminators Stabilizes the GAN Training (ECCV 2022 · similarity 0.86)
- Protecting Intellectual Property of Generative Adversarial Networks From Ambiguity Attacks (CVPR 2021 · similarity 0.86)
- Adversarial Feedback Loop (ICCV 2019 · similarity 0.86)
- Don't let your Discriminator be fooled (ICLR 2019 · similarity 0.86)
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.
