TUT

该数据集名为TUT，包含了大量通过手机拍摄的裂缝图像以及较小规模的互联网来源图像。所有图像均经过手动标注，生成了二值标签，特别关注复杂、噪声大的背景以及复杂的裂缝形状。在TUT数据集中，裂缝像素比例约为3.16%，这样的比例使得模型能够在不过多或少地受到裂缝像素影响的情况下学习到有用的信息。该数据集涵盖了八种不同复杂度的图像场景，其任务是裂缝分割。

This dataset, named TUT, consists of a large number of crack images captured by mobile phones and a smaller-scale set of images sourced from the internet. All images have been manually annotated to generate binary labels, with special focus on complex, high-noise backgrounds and intricate crack shapes. In the TUT dataset, the proportion of crack pixels is approximately 3.16%, a ratio that enables models to learn useful features without being overly influenced by either too many or too few crack pixels. This dataset covers eight image scenarios with varying levels of complexity, and its targeted task is crack segmentation.