# Hi-UCD Series Dataset Hi-UCD (ultra-High Urban Change Detection) series datasets are designed for deep learning based urban semantic change detection. ## News - 2023/9/10 Hi-UCD mini is released. - 2024/7/18 Hi-UCD is released. - **2025/11/01** **We have updated the content of the Hi-UCD training set based on the [issue](https://github.com/Daisy-7/Hi-UCD-S/issues/31). If you downloaded the Hi-UCD dataset before November 1, 2025, please download the patch package from this [link](https://1drv.ms/u/c/0abce178402b585f/EbNqnENEDi5Pl9LeKvMYFcYBfLxkwvhHftQXtuqgha5dRQ?e=Nlr0Hq) to replace the images in the training set.** ## Hi-UCD mini [`Paper`](https://arxiv.org/abs/2011.03247) [`Data`](https://drive.google.com/drive/folders/1fzAn4Bez_S6KX83iYABjAlASCzzhRJPQ?usp=sharing) ![Examples for Hi-UCD mini dataset](./picture/Hi-UCD%20mini.png) The Hi-UCD mini dataset can be downloaded through the Google Drive link. Please submit a download request through the link and I will process your request as soon as possible. ## Hi-UCD [`paper`](https://www.sciencedirect.com/science/article/abs/pii/S0924271622002210) [`Data`](https://docs.google.com/forms/d/e/1FAIpQLSfEJ8z4ubdRbjDjTdr4kFeaLnhtiDNf_lxF-Y6Fee2C-RV9ag/viewform?usp=sf_link) ![Hi-UCD dataset](./picture/Hi-UCD.png) ![public dataset](./picture/public%20dataset.png) The characteristics of Hi-UCD: 1. 40800 pairs ultra-high resolution images (0.1 m) for Tallinn, Estonia. 2. Focus on refined urban semantic change detection. 3. Include 2 years of images, 9 land cover classes and 48 semantic change classes. 4. Tasks that can be performed on this dataset: semantic segmentation, binary change detection, semantic change detection, etc. The Hi-UCD dataset can be downloaded through the [Baidu Drive link](https://pan.baidu.com/s/1Q4wv4jfY4xDHorGSzF0mmw?pwd=0zgp) and [OneDrive link](https://docs.google.com/forms/d/e/1FAIpQLSfEJ8z4ubdRbjDjTdr4kFeaLnhtiDNf_lxF-Y6Fee2C-RV9ag/viewform?usp=sf_link). **If you want to get the test scores, please join our hosted benchmark platform: [semantic change detection](https://www.codabench.org/competitions/3485/).** ## Semantic label and palette |Number|Class|Palette| |---|---|---| |0| unlabeled|(255,255,255)| |1|Water|(0, 153,255 )| |2|grass|(202, 255, 122)| |3|building|(230, 0, 0)| |4|green house| (230, 0, 255)| |5|road|(255, 230, 0)| |6|bridge|(255 ,181 ,197)| |7|others|(0, 255, 230)| |8|bare land|(175, 122, 255)| |9|woodland|(26,255,0)| ## Change label and palette |Number|Class|Palette| |---|---|---| |0| unlabeled|(255,255,255)| |1|unchanged|(0,0,0)| |2|changed|(220, 0, 0)| ## Citation If you use Hi-UCD series dataset in your research, please cite our papers as follows: ``` @article{tian2020hi, title={Hi-UCD: A large-scale dataset for urban semantic change detection in remote sensing imagery}, author={Tian, Shiqi and Ma, Ailong and Zheng, Zhuo and Zhong, Yanfei}, journal={arXiv preprint arXiv:2011.03247}, year={2020} } @article{tian2022large, title={Large-scale deep learning based binary and semantic change detection in ultra high resolution remote sensing imagery: From benchmark datasets to urban application}, author={Tian, Shiqi and Zhong, Yanfei and Zheng, Zhuo and Ma, Ailong and Tan, Xicheng and Zhang, Liangpei}, journal={ISPRS Journal of Photogrammetry and Remote Sensing}, volume={193}, pages={164--186}, year={2022}, publisher={Elsevier} } @article{tian2023temporal, title={Temporal-agnostic change region proposal for semantic change detection}, author={Tian, Shiqi and Tan, Xicheng and Ma, Ailong and Zheng, Zhuo and Zhang, Liangpei and Zhong, Yanfei}, journal={ISPRS Journal of Photogrammetry and Remote Sensing}, volume={204}, pages={306--320}, year={2023}, publisher={Elsevier} } @article{zheng2022changemask, title={ChangeMask: Deep multi-task encoder-transformer-decoder architecture for semantic change detection}, author={Zheng, Zhuo and Zhong, Yanfei and Tian, Shiqi and Ma, Ailong and Zhang, Liangpei}, journal={ISPRS Journal of Photogrammetry and Remote Sensing}, volume={183}, pages={228--239}, year={2022}, publisher={Elsevier} } ``` ## License The use of Hi-UCD series dataset must follow the [licence of open data by Estonian Land Board](https://geoportaal.maaamet.ee/docs/Avaandmed/Licence-of-open-data-of-Estonian-Land-Board.pdf). Hi-UCD series dataset can be used for academic purpose only, **but any commercial use is prohibited**.