Box Supervised Instance Segmentation

Box-supervised instance segmentation aims to improve the efficiency of instance segmentation by using only bounding box annotations instead of more expensive pixel-level masks. Current research focuses on developing novel loss functions and model architectures, including those based on level-set evolution, transformer networks, and affinity-based methods, to accurately predict instance masks from limited box supervision. This approach significantly reduces annotation costs, making large-scale instance segmentation datasets more feasible and impacting various applications, from autonomous driving to remote sensing image analysis. The resulting models often demonstrate comparable performance to fully supervised methods, highlighting the potential of weakly supervised learning in computer vision.