Benchmarking Cross-Domain Few-Shot Object Detection in Aerial Imagery

This paper addresses the challenges of Cross-Domain Few-Shot Object Detection (CD-FSOD), particularly in the context of aerial imagery. CD-FSOD aims to transfer knowledge from data-rich source domains to target domains with sparse labeled data, while confronting the difficulty of adapting to new classes that may differ significantly. To overcome these obstacles, we explore a variety of fine-tuning strategies for object detection models trained on large source domains, with the goal of improving their performance in target domains with few annotations. Additionally, we provide a comprehensive comparison of state-of-the-art object detection models across several cross-domain scenarios. To further facilitate this research, we introduce a novel tool that automates training and testing across diverse configurations, enhancing both reproducibility and efficiency in CD-FSOD studies. Finally, we underscore the importance of selecting appropriate source datasets and fine-tuning strategies to optimize performance in real-world applications. Our code is available here: https://github.com/HichTala/Benchmark_CD-FS-OD_using_HF