Reproduce “The CHASM-SWPC Dataset for Coronal Hole Detection & Analysis” 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
Cutter Beck, Evan Smith, Khagendra Katuwal, Rudra Kafle, Jacob Whitehill
Researchers developed the CHASM-SWPC dataset by digitizing over 50 years of NOAA SWPC expert-drawn coronal hole maps, enabling machine learning models to achieve up to a 17.8% relative improvement in detection accuracy compared to models trained on algorithmic pseudo-labels. This publicly released dataset and tool provide high-quality ground truth for space weather forecasting applications.
Key insights
- High-quality, expert-derived ground truth data significantly improves machine learning model performance, with curated data quality (removing 'bad' annotations) proving more beneficial than raw data quantity.
- A semi-automated annotation pipeline, combining state-of-the-art computer vision (SAM) with human oversight, can efficiently convert extensive historical expert knowledge into machine-readable datasets (1,446 maps in ~14 hours).
- High inter-annotator agreement (mean IoU 0.975-0.988) demonstrates the consistency and reliability of the CHASM tool and the resulting dataset in capturing expert consensus.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.