Source-Free Domain-Adaptive Semi-Supervised Learning for Object Detection in CCTV Images
Hyejin Shin, Gye-Young Kim · 2025
Read Source-Free Domain-Adaptive Semi-Supervised Learning for Object Detection in CCTV Images by Hyejin Shin, Gye-Young Kim (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/pmc-12787354.txt.
Description
Hyejin Shin and Gye-Young Kim built a camera object detector that adapts to new scenes without the source data in a 2025 paper. Detectors trained in one setting fade in real surveillance footage. Privacy law also bars shipping the original training data along. Their method learns from the target scene itself, using few or no labels.
A teacher-student setup merges guesses from mild and strong image tweaks into trustworthy pseudo-labels. A frozen adversarial head, rather than a trained one, keeps the domain pressure steady. A time-shifting weight balances labeled and unlabeled data as training runs. Across city, fog, game, and real camera sets, accuracy rose over seven points on average. Gains held even with only two percent of the target labeled.
Catalog record
| CREATOR | Hyejin Shin, Gye-Young Kim |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 90 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |