RISBench

该数据集名为RISBench，是一个大规模的基准数据集，包含了52,472个图像-语言-标签的三元组，专为参照遥感图像分割（RRSIS）任务而设计。该数据集分为训练集、验证集和测试集，分别包含26,300个、10,013个和16,158个三元组。数据集涵盖26个不同的类别，每个类别都标注了8个属性。图像尺寸为512×512像素，空间分辨率从0.1米到30米不等。该数据集的任务是进行参照遥感图像分割。

This dataset, named RISBench, is a large-scale benchmark dataset containing 52,472 image-language-label triplets, specifically designed for the Referring Remote Sensing Image Segmentation (RRSIS) task. It is split into training, validation and test sets, which contain 26,300, 10,013 and 16,158 triplets respectively. The dataset covers 26 distinct categories, with 8 attributes annotated for each category. All images have a resolution of 512×512 pixels, and their spatial resolutions range from 0.1 meters to 30 meters. The targeted task of this dataset is referring remote sensing image segmentation.

搜集汇总

背景与挑战

背景概述

RISBench是一个大规模视觉语言基准数据集，专为遥感图像分割设计，包含52,472个图像-语言-标签三元组，图像尺寸统一为512x512像素，分辨率范围广泛（0.1m-30m），涵盖26个语义类别，每个类别标注8个属性。