Beyond Few-shot Object Detection: A Detailed Survey

目标检测是计算机视觉中的一个关键领域，专注于准确识别和定位图像或视频中的特定对象。传统的目标检测方法依赖于每个对象类别的大型标记训练数据集，这可能需要耗费大量时间和金钱来收集和注释。为了解决这个问题，研究人员引入了少样本目标检测（FSOD）方法，将少样本学习和目标检测原理相结合。这些方法允许模型仅使用少量标注样本快速适应新的对象类别。虽然传统的FSOD方法以前已经被研究过，但本综述论文全面审查了FSOD研究，特别关注覆盖不同FSOD设置的标准FSOD、广义FSOD、增量FSOD、开放式FSOD和域自适应FSOD。这些方法在减少对大量标记数据集的依赖方面发挥着至关重要的作用，特别是随着需要高效的机器学习模型的需求不断增加。本综述论文旨在提供对上述少样本设置的全面理解，并探讨每个FSOD任务的方法学。它通过详细分析评估协议，全面比较不同FSOD设置的最新方法。此外，它还提供了有关它们的应用、挑战和潜在未来方向的见解，这些都是在有限数据的目标检测领域不断发展的。

Few-shot object detection (FSOD) approaches aim to address the issue of relying on large labeled training datasets for each object category in traditional object detection methods. This paper comprehensively reviews FSOD research across different settings.

The key idea of this paper is to merge few-shot learning and object detection principles in order to allow models to quickly adapt to new object categories with only a few annotated samples. The paper explores different FSOD settings and thoroughly compares state-of-the-art methods across these settings.

The paper offers insights into the applications, challenges, and potential future directions in the evolving field of object detection with limited data. It covers different FSOD settings such as standard FSOD, generalized FSOD, incremental FSOD, open-set FSOD, and domain adaptive FSOD. The paper analyzes state-of-the-art methods in detail based on their evaluation protocols. The experiments are conducted on various datasets and the paper provides open-source code for reproducibility.

Recent related work in this field includes 'Few-shot Object Detection with Attention-RPN and Multi-Relation Detector' and 'Few-shot Object Detection via Feature Reweighting'.