MCG-NJU/VideoChat3-OL617k
收藏资源简介:
VideoChat3-OL617K是VideoChat3使用的在线视频指令数据,旨在训练主动式流媒体视频助手,这些助手能够持续观察传入的视频、积累视觉证据并在适当时刻响应。该数据集将视频-问题-答案三元组转换为因果流监督,首先定位并验证视觉线索间隔,然后转换为带有明确响应状态标记的流序列:`</Silence>`、`</Standby>`和`</Response>`。这些标记教导模型何时保持沉默、继续收集证据和提供答案。本仓库提供JSONL注释文件,原始视频未在此处复制;用户应从下方列出的原始数据集路径解析视频。
VideoChat3-OL617K is the online video instruction data used by VideoChat3. It is designed to train proactive streaming video assistants that continuously observe incoming video, accumulate visual evidence, and respond at the appropriate moment. The dataset converts video-question-answer triples into causal streaming supervision. Visual clue intervals are first localized and verified, then transformed into streaming sequences with explicit response-state tokens: `</Silence>`, `</Standby>`, and `</Response>`. These tokens teach the model when to remain silent, continue collecting evidence, and provide an answer. This repository provides JSONL annotation files. The original videos are not duplicated here; users should resolve videos from the original dataset paths listed below.




