Reproduce “Slender Object Detection: Diagnoses and Improvements” 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
Zhaoyi Wan, Yimin Chen, Sutao Deng, Kunpeng Chen, Cong Yao, Jiebo Luo
Researchers at the University of Rochester and Beihang University systematically diagnose and improve the detection of slender objects, which are frequently overlooked by current object detection systems. Their "self-adaptation" module consistently boosts mean Average Recall for extra slender objects by up to 3.4% across different detector architectures.
Key insights
- A significant performance disparity exists for slender objects, with classification proving more challenging than localization due to increased background within bounding boxes.
- The primary impact of anchors in detection for slender objects lies in their role in label assignment, rather than the regression formulation itself, with anchor-free assignment showing benefits.
- Feature adaptation, particularly through adaptive sampling via Deformable Convolutions, is identified as the critical factor for improving slender object detection, surpassing specialized box representations.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.