ngqtrung/full-modality-video-caption

Full Modality Video Caption Dataset是一个大规模的多模态视频数据集，包含视觉、音频和综合描述。该数据集共有55,940个视频段，每个10秒钟。视频段包括三种类型的描述：视觉描述（由GPT-4o生成）、音频描述（由Qwen3-Omni-30B-A3B-Captioner生成）和综合描述（由Qwen3-Omni-30B-A3B-Instruct生成）。数据集以WebDataset格式提供，包含视频文件和JSON格式的元数据。

The Full Modality Video Caption Dataset is a large-scale multimodal video dataset that includes comprehensive vision, audio, and integrated captions. It contains 55,940 video segments, each 10 seconds long, with three types of captions: vision captions (generated by GPT-4o), audio captions (generated by Qwen3-Omni-30B-A3B-Captioner), and video captions (an integrated multi-modal description generated by Qwen3-Omni-30B-A3B-Instruct). The dataset is provided in the WebDataset format, including video files and metadata in JSON format.