Scribble Annotation

Scribble annotation is a weakly supervised learning approach that uses sparse, user-drawn scribbles instead of dense pixel-level labels to train image segmentation and object detection models. Current research focuses on improving the accuracy and robustness of these models, often employing convolutional neural networks (CNNs), transformers, and diffusion models, along with techniques like consistency regularization and label propagation to overcome the limitations of sparse annotations. This approach significantly reduces the annotation burden, making it particularly valuable for large datasets in medical imaging, remote sensing, and other domains where obtaining dense labels is costly and time-consuming, thereby accelerating progress in these fields.