What question did this study set out to answer?

Develop a few-shot object detection framework capable of identifying novel object categories from only a few annotated bounding box examples.

October 1, 2019

Few-Shot Object Detection via Feature Reweighting

View Full Paper

Ask AI

Bookmark

Authors

BKBingyi KangUniversity of Siedlce ZLZhuang LiuDongbei University of Finance and Economics XWXin WangXihua University

Discussion

Member takes

Implication

Machine learning evaluation demonstrates accurate few-shot novel object detection across multiple benchmark datasets, suggesting effective generalization via episodic feature reweighting.

Key Points

Develop a few-shot object detection framework capable of identifying novel object categories from only a few annotated bounding box examples.
Combined a meta feature learner with a support-driven reweighting module and a prediction module inside a one-stage object detection architecture.
Trained the network end-to-end on base classes with abundant annotations using an episodic few-shot learning scheme and a specialized loss function.
Outperformed well-established baseline models by a substantial margin across multiple few-shot object detection datasets and experimental settings.
Successfully transformed sparse support examples into global importance vectors that dynamically reweighted meta features for novel classes.

Cite This Study

Kang et al. (2019) studied this question.

synapsesocial.com/papers/69df3c3a4fb243fc8e591be7 https://doi.org/10.1109/iccv.2019.00851

View Full Paper

Ask AI

Bookmark

Implication

Machine learning evaluation demonstrates accurate few-shot novel object detection across multiple benchmark datasets, suggesting effective generalization via episodic feature reweighting.

Few-Shot Object Detection via Feature Reweighting

Discussion

Member takes

Implication

Key Points

Cite This Study

Few-Shot Object Detection via Feature Reweighting

Discussion

Member takes

Implication

Key Points

Cite This Study

Also Consider

Also Consider