aifollow.news
返回 Rohan Paul
Rohan Paul· @rohanpaul_ai · X· · 原发布时间 AI 评分54

Perceptron AI 推出 EgoHome,测试 AI 能否读懂家务视频中的手部动作

AI 导读

EgoHome 使用人工标注的家务视频,逐段记录手部位置、任务和动作,并收录超过 750 个短于 500 毫秒的片段;数据、评测指标和评判提示词均已开放。一则相关帖子称,GPT-6 Astra 在该基准测试中排名第一,但材料未给出分数或与其他模型的差距。

原帖引用 · 原文

@rohanpaul_ai

https://t.co/IDzWJcTwbj https://t.co/k7U2E8qcHI

回复Rohan Paul@rohanpaul_ai
GPT-6 Astra came out on top in EgoHome, a new benchmark from Perceptron AI, that tests whether AI can follow what human hands do in chore videos, the skill robots need to learn from people. AI can tell you someone is washing dishes, but not what their hands are doing, and robots need the second part. This kind of benchmarks are so important because Robotics training now happen on large amounts of human video, and AI is used to label those videos automatically (“left hand picks up sponge, right hand turns tap”). If those labels are wrong, the robots learn wrong. https://t.co/t5gL9ZYYmU
在 X 查看这条帖子 ↗
相关上下文Perceptron AI@perceptroninc
Introducing EgoHome: a new open benchmark for dense, sub-second egocentric understanding. Human-annotated household-chore video with 4 annotation types on every clip: hand boxes, subtasks, hand actions, hand captions. It features over 750 segments under 500ms. Data, metrics, and judge prompts are all open. 🧵
在 X 查看这条帖子 ↗

来源:Rohan Paul · x.com

论文
发现内容有误?提交纠错