Ego-OSCAR-550h Stereo-Inertial Egocentric Corpus
Ego-OSCAR论文描述的低成本头戴式双目惯性采集系统及约550小时/相机的人类第一视角语料：1,462个双目session、25名贡献者、40多个室内环境、209,315段自由文本动作标注，并声称逐帧3D手部重建、标定和IMU随数据发布。论文同时声称CAD、固件、采集软件与数据全部开源，但论文源码没有项目、仓库或下载URL，官方GitHub/Hugging Face精确检索也未找到可归属制品，因此当前只登记为高价值待兑现数据，不计作已开放。
- 开放状态
- 项目已核验·制品未开放
- 许可证
- Paper says permissively licensed, but no dataset, hardware or software license file/URL was located
- 商业使用
- not established until the actual release and dataset-specific terms are accessible
- 人工评审结论
- verified_high_value_low_cost_capture_dataset_claim_artifacts_not_located
- 机器人
- human egocentric capture; intended as VLA/world-model pretraining substrate
- 任务
- cooking、dishwashing、sewing and tailoring、cleaning、tool use、multi-step household activities
- 模态
- hardware-synchronized left/right global-shutter RGB、6-axis IMU、free-form action captions、claimed per-frame 3D hand reconstruction、per-session stereo calibration
- 格式
- paper describes 5-minute H.264 MP4 clips, IMU CSV, calibration.json, action labels and hand reconstructions; files not located
- 采集/生成方式
- Dexcin USB stereo camera, Radxa Rock 5C/RK3588, ICM-20948 IMU and Xiao ESP32-S3 in an approximately USD 200/280 g head-mounted rig
- 已知问题
- No release URL, file manifest, checksum, license document, CAD, firmware or software repository located; no pose ground truth/ATE/RPE; 25 contributors in one country; kitchen-heavy distribution; no downstream robot-learning experiment; privacy processing is described but not auditable without files.