← 研踪 PaperTrailarXiv:2608.15652 · v3
Scalable Black-Box Model Attribution for Images
首次提交 2026-08-16 17:45
版本更新 2026-11-17 17:44
雷达观测 2026-11-18 11:33
摘要
The rapid proliferation of generative models raises the model attribution problem: given only an image, can we determine which model produced it? We propose RPA (Raw-Patch Attribution), a lightweight CNN that solves this problem in a strict black box setting. The CNN operates on multiple image patches to handle varying image size and improve accuracy. It attributes more models at higher accuracy than prior work, reaching 98.9% on 25-class DRAGON and 95.0% on 27-class OpenFake; runs in a few milliseconds at a cost nearly independent of candidate-set size; and remains robust to transformations encountered in the wild. Beyond closed-set attribution, its learned representation serves as a reusable fingerprint backbone: it supports open-set rejection and few-shot enrollment, recovers model-family structure without lineage supervision, and groups images from unseen generators by source. Controlled ablations and causal perturbations show that its attribution decisions are driven by a spatially local, low-level signal that behaves as a generator-specific fingerprint.
作者备注:Accepted to WACV 2027. Project page: https://asaf-livne.github.io/RPA/
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v3 · 2026-11-17 17:44 · 最新观测版本
Scalable Black-Box Model Attribution for Images
The rapid proliferation of generative models raises the model attribution problem: given only an image, can we determine which model produced it? We propose RPA (Raw-Patch Attribution), a lightweight CNN that solves this problem in a strict black box setting. The CNN operates on multiple image patches to handle varying image size and improve accuracy. It attributes more models at higher accuracy than prior work, reaching 98.9% on 25-class DRAGON and 95.0% on 27-class OpenFake; runs in a few milliseconds at a cost nearly independent of candidate-set size; and remains robust to transformations encountered in the wild. Beyond closed-set attribution, its learned representation serves as a reusable fingerprint backbone: it supports open-set rejection and few-shot enrollment, recovers model-family structure without lineage supervision, and groups images from unseen generators by source. Controlled ablations and causal perturbations show that its attribution decisions are driven by a spatially local, low-level signal that behaves as a generator-specific fingerprint.
查看此版本解读 →查看此版本原文 ↗