BOOD: Boundary-based Out-Of-Distribution Data Generation
Qilin Liao, Shuo Yang, Bo Zhao, Ping Luo, Hengshuang Zhao · 2025
Read BOOD: Boundary-based Out-Of-Distribution Data Generation by Qilin Liao, Shuo Yang, Bo Zhao, Ping Luo, Hengshuang Zhao (2025) 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-2508-00350.txt.
Description
Qilin Liao and co-authors present BOOD in this 2025 paper, a way to make fake outlier data that helps AI spot unusual inputs. Detectors work better when trained with examples of what does not belong, but good outliers are hard to find. Features near a class boundary carry the most signal, yet locating that boundary is tricky.
BOOD learns a text-guided feature space from the normal data set. It picks the normal features closest to the boundary and nudges them across it. A diffusion model then turns those shifted features into realistic outlier images. On a standard hundred-class benchmark, the method cut false alarms sharply and raised detection scores well past the prior best. The authors say boundary-guided outliers teach cleaner splits between known and unknown data.
Catalog record
| CREATOR | Qilin Liao, Shuo Yang, Bo Zhao, Ping Luo, Hengshuang Zhao |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-00350