Research map / Adversarial robustness and security / Attacks on text and graphs
Protecting Intellectual Property of Generative Adversarial Networks From Ambiguity Attacks
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 reconstructive watermark regularization | DCGAN · CIFAR-10 and CUB-200 | FID: 26.54 ± 1.04 -> 24.83 ± 0.37; 58.34 ± 1.50 -> 53.07 ± 4.07 | Helped |
| Add reconstructive watermark regularization and sign loss | DCGAN · CIFAR-10 and CUB-200 | FID: 26.54 ± 1.04 -> 26.27 ± 0.54; 58.34 ± 1.50 -> 56.64 ± 2.74 | Helped |
| Add reconstructive watermark regularization and sign loss | SRGAN · Set5, Set14, BSD100 | PSNR/SSIM: 29.38/0.85 -> 29.14/0.85; 25.92/0.71 -> 26.00/0.72; 25.08/0.67 -> 25.35/0.67 | Mixed |
| Add reconstructive watermark regularization | CycleGAN · Cityscapes label → photo | per-pixel acc.: 0.55 -> 0.55 | No clear effect |
| Add sign loss to the protected CycleGAN | CycleGANw · Cityscapes label → photo | per-pixel acc.: 0.55 -> 0.58 | Helped |
| Increase λ for reconstructive watermark regularization (ablation) | DCGANw · CIFAR-10 | FID: 25.88 -> 47.38 | Hurt |
| Set trigger constant c to 0 (ablation) | DCGANw · CIFAR-10 | FID: 26.05 -> 276.36 | Hurt |
| Fine-tune the protected DCGAN without watermark regularizers (ablation) | DCGANws · CIFAR-10 | FID: 26.54 ± 1.04 -> 30.50 ± 1.10 | Hurt |
| Overwrite the watermark in the protected DCGAN (ablation) | DCGANws · CIFAR-10 | FID: 26.54 ± 1.04 -> 35.68 ± 1.10 | Hurt |
| Modify 10% of the embedded scale signs (ablation) | DCGANws · CIFAR-10 | FID: 26 -> 91 | Hurt |
| Fine-tune the protected SRGAN without watermark regularizers (ablation) | SRGANws · Set5, Set14, BSD100 | PSNR/SSIM: 29.14/0.85 -> 26.07/0.85; 26.00/0.72 -> 23.75/0.72; 25.35/0.67 -> 23.58/0.68 | Hurt |
| Overwrite the watermark in the protected SRGAN (ablation) | SRGANws · Set5, Set14, BSD100 | PSNR/SSIM: 29.14/0.85 -> 27.65/0.84; 26.00/0.72 -> 25.08/0.72; 25.35/0.67 -> 24.66/0.68 | Hurt |
Closest papers by meaning
- Watermark-embedded Adversarial Examples for Copyright Protection against Diffusion Models (CVPR 2024 · similarity 0.87)
- IPRemover: A Generative Model Inversion Attack against Deep Neural Network Fingerprinting and Watermarking (AAAI 2024 · similarity 0.86)
- Tracking the Copyright of Large Vision-Language Models through Parameter Learning Adversarial Images (ICLR 2025 · similarity 0.85)
- Adversarial Perturbations Cannot Reliably Protect Artists From Generative AI (ICLR 2025 · similarity 0.84)
- Vision-Language Model IP Protection via Prompt-based Learning (CVPR 2025 · similarity 0.84)
- MAP: MAsk-Pruning for Source-Free Model Intellectual Property Protection (CVPR 2024 · similarity 0.84)
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.