Reproduce “OSAD: Open-Set Aircraft Detection in SAR Images” 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
Xiayang Xiao, Zhuoxuan Li, Haipeng Wang
Researchers at Fudan University developed OSAD, an open-set aircraft detection framework for SAR images, that successfully identifies known aircraft while distinguishing unknown objects. The method demonstrated superior detection of unknown objects (e.g., up to 18.36% AP_unknown) and preserved high accuracy for known categories on a new OS-SAR-Aircraft dataset.
Key insights
- Robust open-set detection in SAR requires a combined approach that enhances feature learning, generates unbiased proposals, and employs a discriminative classification mechanism.
- Decoupling the Region Proposal Network's classification from localization via quality-driven pseudo-label generation improves its ability to generalize to unknown objects.
- Prototype-based contrastive learning, utilizing dynamically updated class prototypes, effectively creates a separable feature space for known and unknown categories, reducing misclassification.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.