tao-generate-video-reasoning-annotations

【数据与研究】Multi-step video annotation pipeline that turns raw videos into Chain-of-Thought training data — multi-level captions, structured descriptions, and QA pairs (MCQ, binary, open-ended) with reasoning traces, via VLM/LLM distillation. Use when the user wants to "create video training data", "generate video QA datasets", "build CoT reasoning traces from …

独立 GitHub 快照 · 自动分类 · 以原作者文档为准