Semi-supervised Source Detection in Astronomical Images: New Benchmark and Strong Baseline
Researchers have introduced a new benchmark and a novel semi-supervised learning framework for detecting sources in astronomical images. The benchmark, LAMOST-DET, includes over 18,000 images and nearly 730,000 source instances, addressing the scarcity of annotated astronomical data. Their framework, Nova Teacher, integrates several modules to effectively detect dense sources even with limited annotations, showing significant improvements in mean Average Precision (mAP) over existing methods. AI
IMPACT Provides a new dataset and improved methodology for AI-driven analysis in astronomy, potentially accelerating discoveries.