stonkens/droid_iris_full

droid_iris_full 是从 lerobot/droid_1.0.1 数据集中通过实验室过滤得到的子集，使用基于 aggregated-annotations-030724.json 的多数投票 collector_id 到 lab 的映射，筛选出 lab 为 IRIS 的数据。该数据集包含 7665 个 episodes、2,061,097 帧和 4437 个任务，基于 LeRobot v3.0 代码库。它仅包含轻量级 parquet 数据（如本体感觉、动作、语言和元数据），不包括原始 DROID 视频（约 200 GB 的 AV1 MP4 格式），且已从 info.json 中移除 observation.images.* 视频特征。droid_iris_full is a lab-filtered subset of the lerobot/droid_1.0.1 dataset, filtered using a majority-vote collector_id to lab mapping from aggregated-annotations-030724.json to select data where lab = IRIS. It includes 7665 episodes, 2,061,097 frames, and 4437 tasks, based on the LeRobot v3.0 codebase. The subset contains only lightweight parquet data (e.g., proprio, actions, language, metadata), excluding the original DROID videos (approximately 200 GB of AV1 MP4 format), and the observation.images.* video features have been dropped from info.json.