Research map / Image generation / Text-to-image diffusion
Efficient, Robust, and Anti-Collusion Fingerprinting of Image Diffusion Models
Jianwei Fei, Yunshu Dai, Zhihua Xia, Xiaochun Cao, Jiantao Zhou, Alessandro Piva, Benedetta Tondi
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 anti-collusion transformation (ACT) to the PNM (ablation) | Ours · COCO, 2-party collusion | FID of colluded model: 23.55 -> 79.51 | Improved |
| add worst-case regularization loss L_reg (ablation) | Ours · COCO fine-tuning attack | Bit Acc after 10,000 fine-tuning steps: above 90%, approximately 7% lower without L_reg | Improved |
| raise fingerprint loss weight lambda_m from 0.1 to 10 (ablation) | Ours · COCO (generation) | PSNR_b: 31.95 -> 30.92 | Hurt |
| raise fingerprint loss weight lambda_m from 0.1 to 10 (ablation) | Ours · COCO (generation) | Bit Acc under JPEG (%): 90.37 -> 94.65 | Improved |
| raise fingerprint loss weight lambda_m from 0.1 to 10 (ablation) | Ours · COCO (generation) | Bit Acc no attack (%): 99.45 -> 99.85 | Improved |
| add noise-layer augmentation during fine-tuning (ablation) | Ours · COCO image-level attacks | Bit Acc | Improved |
| add Cheng compression model as augmentation (ablation) | Ours · COCO purification attack | Bit Acc: over 10% increase in Bit Acc | Improved |
| replace EfficientNet-B0 fingerprint decoder with ResNet18 (ablation) | Ours · COCO (generation) | Bit Acc under JPEG (%): 92.04 -> 91.20 | No clear effect |
| replace EfficientNet-B0 fingerprint decoder with ResNet50 (ablation) | Ours · COCO (generation) | Bit Acc under JPEG (%): 92.04 -> 91.07 | No clear effect |
| remove PNM at inference (structural attack) | Ours · COCO | FID: 24.03 -> 185.22 | Hurt |
| reinitialize PNM weights (structural attack) | Ours · COCO | Bit Acc (%): 99.57 -> 50.30 | Hurt |
| increase number of colluders N from 3 to 20 | Ours with ACT · COCO multi-party collusion | PSNR / LPIPS of colluded model: 12.64/0.59 -> 10.96/0.70 | No clear effect |
Closest papers by meaning
- WOUAF: Weight Modulation for User Attribution and Fingerprinting in Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.89)
- Fingerprinting Denoising Diffusion Probabilistic Models (CVPR 2025 · similarity 0.88)
- SleeperMark: Towards Robust Watermark against Fine-Tuning Text-to-image Diffusion Models (CVPR 2025 · similarity 0.85)
- SIF: Semantically In-Distribution Fingerprints for Large Vision-Language Models (CVPR 2026 · similarity 0.85)
- OmniMark: Efficient and Scalable Latent Diffusion Model Fingerprinting (AAAI 2025 · similarity 0.85)
- ManiFPT: Defining and Analyzing Fingerprints of Generative Models (CVPR 2024 · similarity 0.85)
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.
