InSpaceType
收藏资源简介:
InSpaceType数据集由南加州大学和加州大学洛杉矶分校的研究团队创建,专注于室内单目深度估计,包含1260个高质量的RGBD对,覆盖多种室内场景类型。数据集通过高分辨率立体相机ZED-2i采集,确保了数据的密集性和高分辨率,适用于现代应用如机器人导航和AR/VR感知。数据集的创建旨在解决现有数据集在不同空间类型上性能不均的问题,特别是在罕见或不常见空间类型上的表现。该数据集的应用领域包括家庭自动化、机器人导航和AR/VR环境感知,旨在提高深度估计模型在实际应用中的鲁棒性和泛化能力。
The InSpaceType dataset was constructed by research teams from the University of Southern California and the University of California, Los Angeles, with a core focus on indoor monocular depth estimation. It comprises 1,260 high-quality RGBD pairs covering a broad spectrum of indoor scene types. Collected using the high-resolution stereo camera ZED-2i, the dataset ensures data density and high resolution, making it suitable for modern applications such as robotic navigation and AR/VR perception. This dataset was developed to address the uneven performance of existing datasets across different spatial types, particularly their poor performance on rare or uncommon spatial categories. Its application domains include home automation, robotic navigation, and AR/VR environmental perception, with the goal of improving the robustness and generalization capability of depth estimation models in real-world applications.




