Analyzing the Impact of Low-Rank Adaptation for Cross-Domain Few-Shot Object Detection in Aerial Images
Hicham TALAOUBRID, Anissa MOKRAOUI, Ismail BEN AYED, Axel PROUVOST, Sonimith HANG, Monit KORN, Rémi HARVEY · 2025
Read Analyzing the Impact of Low-Rank Adaptation for Cross-Domain Few-Shot Object Detection in Aerial Images by Hicham TALAOUBRID, Anissa MOKRAOUI, Ismail BEN AYED, Axel PROUVOST, Sonimith HANG, Monit KORN, Rémi HARVEY (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-2504-06330.txt.
Description
Hicham Talaoubrid and co-authors published a 2025 paper on teaching small AI models to spot objects in aerial photos with very few examples. Models often must move to a new region or task where labeled photos are scarce. A method called LoRA adapts big models cheaply by training only a small add-on part. The authors test whether it also helps small models in this setting.
They added LoRA to a detection model and tested it on two aerial datasets of planes, ships, and ground objects. When the model was first fine-tuned normally and LoRA was applied after, results improved slightly in the hardest settings, with one or five examples per class. With more examples, plain full fine-tuning still won. The authors read this as a sign that light adaptation methods deserve more study for few-shot learning. They share their code for others to build on.
Catalog record
| CREATOR | Hicham TALAOUBRID, Anissa MOKRAOUI, Ismail BEN AYED, Axel PROUVOST, Sonimith HANG, Monit KORN, Rémi HARVEY |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 22 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-06330