kshitijrajsharma/pangaea2-vhr

PANGAEA-2 VHR数据集是一个集合，包含四个独立的高分辨率遥感数据集：PureForest、xView2、mpv4ger和SN3 Roads。每个数据集针对不同的地理空间任务：PureForest用于13类树种分类，包含135,569个样本，图像尺寸为250x250，具有4个波段（近红外、红、绿、蓝）；xView2用于建筑物损坏分割（0-4级），包含约10,101对图像，尺寸为1024x1024，RGB波段；mpv4ger用于二元太阳能电池板分类，包含13,812个样本，图像尺寸为320x320，RGB波段；SN3 Roads用于道路网络提取，包含3,708个样本，图像尺寸为1300x1300，RGB波段（uint16）。所有数据集以Parquet格式存储，并保留了原始地理坐标信息（如CRS）。该数据集旨在为遥感研究提供基准数据，支持图像分类、分割、目标检测和变化检测等任务。The PANGAEA-2 VHR dataset is a collection of four independent very-high-resolution remote sensing datasets: PureForest, xView2, mpv4ger, and SN3 Roads. Each dataset targets different geospatial tasks: PureForest is for 13-class tree species classification with 135,569 samples, image size 250x250, and 4 bands (NIR, R, G, B); xView2 is for building damage segmentation (levels 0-4) with approximately 10,101 image pairs, size 1024x1024, RGB bands; mpv4ger is for binary solar panel classification with 13,812 samples, image size 320x320, RGB bands; SN3 Roads is for road network extraction with 3,708 samples, image size 1300x1300, RGB bands (uint16). All datasets are stored in Parquet format and retain original geospatial information (e.g., CRS). This dataset serves as a benchmark for remote sensing research, supporting tasks such as image classification, segmentation, object detection, and change detection.