Voxel51/lingbot-depth-subset
收藏资源简介:
这是一个名为lingbot-depth-subset的数据集,它是robbyant/mdm_depth(LingBot-Depth训练语料库)的一个代表性重采样子集,已重新格式化为FiftyOne的分组数据集结构,其中gt_depth和raw_depth字段作为fo.Heatmap。数据集包含13,149个样本(10,207个组),涵盖三个子集合:RobbyReal(真实世界室内RGB-D捕获)、RobbyVla(真实世界VLA机器人操作,使用Franka机械臂)和RobbySim(模拟渲染)。该子集主要用于验证FiftyOne导入管道,并支持对数据集结构、格式和已知问题的探索性分析,而无需下载完整的多TB原始数据集。它适用于深度估计和深度完成任务,但并非用于重现原始论文训练结果的均匀随机样本。数据集语言为英语(仅图像和深度传感器数据,无自然语言内容),许可证为apache-2.0,但需注意原始数据许可证可能存在不一致性。
This is a dataset named lingbot-depth-subset, which is a representative, recon-sampled subset of robbyant/mdm_depth (the LingBot-Depth training corpus), reformatted into FiftyOnes grouped-dataset structure with gt_depth and raw_depth as fo.Heatmap fields. The dataset contains 13,149 samples (10,207 groups) spanning three sub-collections: RobbyReal (real-world indoor RGB-D captures), RobbyVla (real-world VLA robot manipulation using a Franka arm), and RobbySim (simulated renders). This subset is primarily intended for validating FiftyOne import tooling and enabling exploratory analysis of the datasets structure, formats, and known quirks without downloading the full multi-terabyte source dataset. It is suitable for depth estimation and depth completion tasks but is not a uniformly random sample for reproducing the papers training results. The dataset language is English (image and depth-sensor data only, no natural language content), with an apache-2.0 license, though note potential inconsistencies in the original data license.




