Reproduce “Coreset Selection for Object Detection” 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
Hojun Lee, Suyoung Kim, Junhoo Lee, Jaeyoung Yoo, Nojun Kwak
Introduces Coreset Selection for Object Detection (CSOD), a method that addresses the multi-object nature of real-world datasets by using imagewise-classwise feature representation and submodular optimization. CSOD achieves a +6.4%p AP50 improvement over random selection on Pascal VOC and shows consistent gains across BDD100k and COCO2017 while maintaining cross-architecture effectiveness.
Key insights
- The 'imagewise-classwise vector' provides an effective and compact representation of multi-object information within an image, crucial for adapting coreset selection to object detection.
- Submodular optimization, carefully balancing representativeness and diversity through a tunable hyperparameter 'λ', is essential for selecting a high-quality coreset for complex, multi-object images.
- Coreset selection for object detection requires a fundamental shift from object-level to image-level selection while retaining class-specific information, as traditional classification-focused methods are inadequate.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.