Map Border Dataset was created in order to facilitate training and evaluation of methods for detection and segmentation tasks on historical cadastral maps. The data were collected from materials owned by the State Administration of Land Surveying and Cadastre of the Czech Replublic. We have manually created ground-truths for four tasks: Landmark Detection, Cadastre Border Detection, Edgeline Detection and Breakpoint Detection. The dataset is freely available for education and research purposes. However, any other use is strictly excluded!
The dataset consists of 100 map sheets with resolution of circa 8400 x 6850 pixels. It is divided into training, validation and test parts containing 70, 10 and 20 images respectivelly. The ground-truths for each task are provided as binary masks where white colour indicates the objects of interest. The ground truths for the border and ladmark detection tasks are provided for all dataset parts. However, the other two types of ground-truths were created only for the test part since we haven't trained any neural models for the tasks of Edgeline and Breakpoint detection. The ground-truths were thus utilized only for evaluation purposes.
The original map sheets are stored as .JPG images. The ground-truth masks are provided in the .png format.
This dataset is licensed under the Attribution-NonCommercial-ShareAlike 4.0 International, so commercial use in any form is excluded.
For further information about this dataset, please, see the paper below:
Please, cite this paper when you use this dataset in your experiments.