AI Paper Atlas

Research map / Image generation / Text-to-image diffusion

Scalable Dual Fingerprinting for Hierarchical Attribution of Text-to-Image Models

Jianwei Fei, Yunshu Dai, Peipeng Yu, Zhe Kong, Jiantao Zhou, Zhihua Xia

ICCV 2025 · Image generation › Text-to-image diffusion · approach: Fairness and bias in generative models · 1 citations (OpenAlex)

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 changedWhereMeasuredEffect
add attack simulation (noise) layer during fine-tuning (ablation)FI-LoRA · JPEG compression or Gaussian noise (Std=0.2)Bit Acc: 50% -> nearly 100%Improved
use FI-LoRA instead of WOUAF for image fidelitySD2-base fingerprinting · COCOPSNR_b: 31.345 -> 32.359Improved
use FI-LoRA instead of baselines for generation qualitySD2-base fingerprinting · COCOFID_r: 23.726 -> 23.272Improved
embed two fingerprints (64 bits total) instead of oneFI-LoRA · COCOBit Acc (%): 99.67 (Per. Norm.) -> 99.96 / 99.91No clear effect
apply 50% pruning or noise std 0.05 model attackFI-LoRA · model modification attacksBit Acc: about 5% below Per. Norm.Hurt
adversarial fine-tuning of the model for 10k stepsFI-LoRA · model fine-tuning attackBit Acc: remains above 80% vs below 70% for baselinesMixed
use a single 64-bit fingerprint instead of two 32-bit fingerprintsFI-LoRA · capacity limitconvergenceHurt
2x brightness attack on generated imagesFI-LoRA · image-level attacksBit Acc: below 80% with PSNR 10 or lowerHurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy