Cos R-CNN for Online Few-shot Object Detection

CoRR(2023)

引用 0|浏览9
暂无评分
摘要
We propose Cos R-CNN, a simple exemplar-based R-CNN formulation that is designed for online few-shot object detection. That is, it is able to localise and classify novel object categories in images with few examples without fine-tuning. Cos R-CNN frames detection as a learning-to-compare task: unseen classes are represented as exemplar images, and objects are detected based on their similarity to these exemplars. The cosine-based classification head allows for dynamic adaptation of classification parameters to the exemplar embedding, and encourages the clustering of similar classes in embedding space without the need for manual tuning of distance-metric hyperparameters. This simple formulation achieves best results on the recently proposed 5-way ImageNet few-shot detection benchmark, beating the online 1/5/10-shot scenarios by more than 8/3/1%, as well as performing up to 20% better in online 20-way few-shot VOC across all shots on novel classes.
更多
查看译文
关键词
detection,r-cnn,few-shot
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要