数据集:
sjyhne/mapai_dataset
许可:
cc-by-4.0The dataset comprises 7500 training images and 1500 validation images from Denmark. The test dataset is split into two tasks, where the first task (1368 images) is to segment the buildings only using aerial images. In contrast, the second task (978 images) allows using aerial images and lidar data. All data samples have a resolution of 500x500. The aerial images are RGB images, while the lidar data are rasterized. The ground truth masks have two classes, building, and background. All data derives from a production setting, which means that there will be buildings that are not present in the ground truth and vice versa.
The MapAI Dataset has four splits; train , validation , task1_test , task2_test . Below are the statistics for each split.
Dataset Split | Number of Instances in Split |
---|---|
Train | 7 500 |
Validation | 1 500 |
Task1_test | 1 368 |
Task2_test | 978 |
The purpose of the dataset is to help develop models for accurate segmentation of buildings, which will help downstream tasks such as 3-dimensional building construction.
@article{Jyhne2022, author = {Sander Jyhne and Morten Goodwin and Per-Arne Andersen and Ivar Oveland and Alexander Salveson Nossum and Karianne Ormseth and Mathilde Ørstavik and Andrew C Flatman}, doi = {10.5617/NMI.9849}, issn = {2703-9196}, issue = {3}, journal = {Nordic Machine Intelligence}, keywords = {Aerial Images,Deep Learning,Image segmentation,machine learning,remote sensing,semantic segmentation}, month = {9}, pages = {1-3}, title = {MapAI: Precision in Building Segmentation}, volume = {2}, url = {https://journals.uio.no/NMI/article/view/9849}, year = {2022}, }