Epistemic Uncertainty for Generated Image Detection
Jun Nie, Yonggang Zhang, Tongliang Liu, Yiu-ming Cheung, Bo Han, Xinmei Tian · 2024
Read Epistemic Uncertainty for Generated Image Detection by Jun Nie, Yonggang Zhang, Tongliang Liu, Yiu-ming Cheung, Bo Han, Xinmei Tian (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-05897.txt.
Description
Jun Nie and co-authors present a 2024 computing paper on spotting AI-generated images. As image generators improve, telling real photos from fake ones becomes a security concern. Their key insight links the task to how unsure a model is about an image.
Models trained on natural images tend to show high epistemic uncertainty, a measure of what the model does not know, when shown generated images. The shift between real and fake data shows up clearly in that uncertainty space. The team uses large pre-trained vision models to estimate the uncertainty and flags high scores as generated. Wide tests support the approach. The method turns a detection problem into an uncertainty problem, resting on the strength of big pre-trained models.
Catalog record
| CREATOR | Jun Nie, Yonggang Zhang, Tongliang Liu, Yiu-ming Cheung, Bo Han, Xinmei Tian |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 89 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-05897