Scribble Supervision

Scribble supervision is a weakly supervised learning approach for image segmentation that uses sparse, hand-drawn annotations (scribbles) instead of dense pixel-level labels, significantly reducing annotation costs. Current research focuses on developing novel loss functions and model architectures, such as dual-branch networks and transformer-based methods, to effectively leverage these sparse scribbles for accurate segmentation, often incorporating techniques like pseudo-label generation and consistency regularization. This approach holds significant promise for various applications, particularly in medical image analysis where obtaining fully annotated datasets is challenging, enabling the training of high-performing segmentation models with substantially reduced annotation effort.