Early High-Frequency Injection for Geometry-Sensitive OOD Detection
Chuanjie Cheng, Ningkang Peng, Chenxi Liu, Yifan He, Peirong Ma, Yanhui Gu · 2026
Read Early High-Frequency Injection for Geometry-Sensitive OOD Detection by Chuanjie Cheng, Ningkang Peng, Chenxi Liu, Yifan He, Peirong Ma, Yanhui Gu (2026) 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-2605-20728.txt.
Description
Chuanjie Cheng and five colleagues propose early high-frequency injection for geometry-sensitive out-of-distribution detection. Post-hoc detectors grade logits or features after training, so success depends on geometry already encoded in representations. Their band-wise analysis finds low bands give weaker in-versus-out gaps, while high bands separate better. Their EIHF method exposes high-band evidence before the first convolution without shifting the training goal, improving CIFAR-100 results and ImageNet-100 scores.
Catalog record
| CREATOR | Chuanjie Cheng, Ningkang Peng, Chenxi Liu, Yifan He, Peirong Ma, Yanhui Gu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 37824 |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.20728