Diffuse to Detect: Bi-Level Sample Rebalancing with Pseudo-Label Diffusion for Point-Supervised Infrared Small-Target Detection
Researchers have developed a new framework called Diffuse to Detect to improve infrared small-target detection using point-based supervision. This method addresses challenges like unstable pseudo-labeling in complex imagery and imbalanced sample distributions. By using a physics-induced annotation strategy based on heat diffusion, the system generates more reliable pseudo-masks from single-point labels and employs a bi-level dual-update framework to optimize detector and sample weights. AI
IMPACT Introduces a novel approach to enhance the accuracy and efficiency of infrared small-target detection systems.