Reproduce “Query Adaptive Few-Shot Object Detection with Heterogeneous Graph Convolutional Networks” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Guangxing Han, Yicheng He, Shiyuan Huang, Jiawei Ma, Shih-Fu Chang

A novel few-shot object detection framework, QA-FewDet, employs heterogeneous Graph Convolutional Networks to model diverse relationships among object classes and region proposals, facilitating robust detection with limited examples. The method consistently outperforms state-of-the-art models on PASCAL VOC and MSCOCO datasets, demonstrating significant accuracy improvements, such as over 4.0% AP50 on PASCAL VOC compared to its baseline.

Key insights

  • Modeling diverse relationships via heterogeneous GCNs, particularly the mutual adaptation between class prototypes and query proposals, is critical for achieving robust few-shot object detection.
  • Leveraging inter-class relationships, including those with base classes, significantly enhances the robustness and representativeness of novel class prototypes.
  • Both local contextual information from overlapping proposals and global scene context from the entire query image contribute positively to refining object proposal features for improved detection accuracy.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.