Reproduce “BandRe: Rethinking Band-Pass Filters for Scale-Wise Object Detection Evaluation” 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
Yosuke Shinya
This paper introduces BandASAP, a new metric for evaluating object detection performance across different object scales, by reframing scale-wise evaluation using band-pass filters from signal processing and fuzzy set theory. BandASAP uses overlapping triangular filters to provide more nuanced and reliable insights into model strengths and weaknesses at specific scales compared to existing coarse-grained or highly sensitive fine-grained metrics.
Key insights
- Reframing scale-wise evaluation using band-pass filters with non-binary membership functions (e.g., triangular) offers a more nuanced assessment than traditional rectangular filters.
- Overlapping triangular filters increase the reliability of scale-wise metrics, particularly for scales with sparse object representations, by allowing objects to contribute to multiple adjacent scale ranges.
- This approach provides a more continuous and intuitive understanding of object detector performance across the entire scale spectrum.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.