In order to better develop and evaluate multi-level map generation methods. We collect a large-scale SI based Multi-Level Map Generation (MLMG) dataset with equal amounts of data at four levels (15, 16, 17, and 18) to ensure that all levels are balanced for training. To keep the diversity of data, the dataset is collected from the United States and China, two countries with large differences in SI image and map image. For each country, we provide 8000 samples for training and 1700 samples for evaluation. Each sample consists of an SI image, a corresponding map image, a level label, and an element label.
The MLMG dataset is now publicly available!
link: https://pan.baidu.com/s/1i8Z6c6rVLWEnU8RY1vANFQ
code: zpk2
