Bridging Day and Night: Target-Class Hallucination Suppression in Unpaired Image Translation
Shuwei Li, Lei Tan, Robby T. Tan · 2026
Read Bridging Day and Night: Target-Class Hallucination Suppression in Unpaired Image Translation by Shuwei Li, Lei Tan, Robby T. Tan (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-2602-15383.txt.
Description
Li and colleagues tackle day-to-night image mapping, downstream but hard given big shifts and no pixel-level guidance. Current tools often fake objects from target classes such as traffic signs, vehicles, and man-made light glows, harming scores.
The team offers a setup spotting and blocking ghosts of target traits. To spot ghosts, they build a dual-head judge also doing meaning splitting to spot ghost stuff. To block ghosts, they offer class samples, made from labeled target-domain objects, which act as pins for each class. Built on a Schrodinger Bridge model, the setup stepwise tunes, moving spotted ghost traits away from samples, keeping object meaning across the path. Tests show the method beats current routes. On BDD100K, it lifts mAP by 15 and a half percent for day-to-night shift. It shows about 32 percent rise for classes such as traffic lamps apt to ghosts.
Catalog record
| CREATOR | Shuwei Li, Lei Tan, Robby T. Tan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-15383