Skip to page content
SAVRN
Search Plan a project

Independent publisher

Sanbei101

Sanbei101

Models in Library0
Datasets in Library1
Models on Hugging Face1
Followers—

Compare the published evidence

This profile connects the 1 model and dataset record shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.

Published tasks: Image and text to text.

Reported licenses: other. Open a record to review its specific license and conditions.

Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.

Datasets

Dataset · Image and text to text

dexforge-episode-smolvlm-demo

Sanbei101

单条双臂遥操作 MCAP 录制切分出的 6 个 Episode 及抽帧结果,用于在 Colab 上验证 SmolVLM2-500M-Video-Instruct 能否替代 LLM 网关做 Episode 语义预标注。 含 /fsmstate 状态机、双臂末端位姿、夹爪关节、XR 手柄输入等信号。 - 原始录制:recording/rec202609141354110.mcap(约 36.4 秒,219935 条消息) - 切分口径:平台现有规则 episode.Method = fsmposev1,即优先按 /fsmstate 非零活动区间切分;本录制 6 个 Episode 全部命中 fsm 边界,置信度 0.90 - 帧抽取:每个 Episode 内按 frameCount 预算在 3 路视频流(左右彩相机 + XR 相机) 上均匀取样,JPEG 落盘 frames/index.json 每条形如: 帧按时间顺序编号。一个 Episode 取 8 帧,三路相机混合取样 (例如 episode-000 是 3 帧左相机 + 3 帧右相机 + 2 帧 XR 相机)。 平台 internal/pipelines/semanticepisodetag 活动当前把同样的证据交给 OpenAI 兼容视觉模型,产出 actionId / task / caption / outcome / confidence / needsHumanReview / evidenceFrameIds。这个数据集就是那条链路的输入快照, 用来先在 Colab…

Publicly accessible other n<1K