CSeg
收藏资源简介:
CSeg数据集是由休斯顿大学创建的,专门用于文本提示驱动的变化检测任务。该数据集包含100,311对未对齐的图像,涵盖室内、室外、街道、合成和卫星图像等多种场景。数据集通过图像修复技术生成真实的变化对,避免了手动标注的繁琐过程,并引入了“红鲱鱼”变化以增强模型的泛化能力。CSeg数据集的应用领域广泛,包括情境感知、基础设施评估、环境监测和工业自动化等,旨在解决传统变化检测方法在处理未对齐图像时的局限性。
The CSeg dataset was developed by the University of Houston, specifically tailored for text-prompt-driven change detection tasks. It contains 100,311 unaligned image pairs covering diverse scenarios including indoor, outdoor, street, synthetic, and satellite images. Realistic change pairs are generated via image inpainting techniques, eliminating the tedious manual annotation process, and introduces "red herring" changes to enhance the model's generalization ability. The CSeg dataset has a wide range of application domains such as context awareness, infrastructure assessment, environmental monitoring, and industrial automation, aiming to address the limitations of traditional change detection methods when handling unaligned images.




