Separate and Conquer: Cluster Infrared Small Target Label Generation With Boundary and Direction Sensitivity (opens in new tab)

Cluster infrared small target detection (IRSTD) aims to extract densely distributed weak targets from complex backgrounds. Existing deep learning approaches still heavily rely on large-scale fully annotated masks. Although current label generation techniques significantly improve annotation efficiency, they often lead to label adhesion and fail to accurately distinguish adjacent targets. To address this, we propose BDSNet, the first label generation framework for cluster IRSTD, which incorpor...