IRS

IRS数据集是由香港浸会大学计算机科学系创建的大规模合成室内机器人立体数据集，包含超过10万对960x540分辨率的立体RGB图像及其高精度的表面法线和视差地图。数据集利用定制的Unreal Engine 4插件，通过先进的渲染技术模拟真实世界的光影效果和材质，覆盖多种室内场景类型和视觉属性，如亮度变化、光反射/传输、镜头光晕等。IRS数据集特别适用于训练深度模型进行视差和表面法线估计，旨在解决室内机器人定位、导航和交互中的场景理解和重建问题。

The IRS Dataset is a large-scale synthetic indoor robot stereo dataset created by the Department of Computer Science of Hong Kong Baptist University. It contains over 100,000 pairs of stereo RGB images with a resolution of 960×540, along with high-precision surface normal maps and disparity maps. The dataset utilizes a custom Unreal Engine 4 plugin and advanced rendering technologies to simulate real-world lighting effects and material properties, covering diverse indoor scene types and visual attributes including brightness variations, light reflection/transmission, and lens flares. The IRS Dataset is specifically designed for training deep models in disparity and surface normal estimation, with the goal of addressing scene understanding and reconstruction challenges in indoor robot localization, navigation, and interaction.