VGGSound
收藏资源简介:
VGGSound数据集是一个用于视频引导音频生成的工业级标准数据集,涵盖了电影、游戏和商业广告等多种真实场景。该数据集包含约173,000个视频片段及其对应的音频,旨在为视频到音频生成任务提供高质量的训练数据。数据集的创建过程包括视频分类、内容下载或录制、音频重采样和单声道转换、自动化标注和复杂视频理解等步骤,确保了数据的高质量和多样性。该数据集主要应用于音频生成领域,特别是在少样本设置下生成高质量同步音频,解决视频背景音缺失的问题。
The VGGSound dataset is an industry-standard dataset for video-guided audio generation, covering diverse real-world scenarios including movies, games, and commercial advertisements. It contains approximately 173,000 video clips and their corresponding audios, aiming to provide high-quality training data for video-to-audio generation tasks. The dataset creation process includes steps such as video classification, content downloading or recording, audio resampling and mono conversion, automated annotation, and complex video understanding, which ensures the high quality and diversity of the data. It is primarily applied in the field of audio generation, particularly for generating high-quality synchronized audio under few-shot settings to address the issue of missing background audio in videos.




