DVMSet
收藏资源简介:
DVMSet是由中国科学院自动化研究所和腾讯公司联合构建的一个多样化的视频-音乐数据集,涵盖了多种场景,如宣传视频、广告、动画和短视频等。该数据集包含3839个高质量的视频-音乐对,通过去除语音轨道和手动筛选来确保数据质量。数据集的创建过程包括多阶段的数据收集和处理,旨在为视频与音乐的语义和节奏对齐研究提供支持。DVMSet主要应用于视频背景音乐生成领域,旨在解决视频与音乐自动对齐的问题,提升视频内容的吸引力和观众的沉浸感。
DVMSet is a diverse video-music dataset jointly constructed by the Institute of Automation, Chinese Academy of Sciences and Tencent. It covers various scenarios including promotional videos, advertisements, animations and short videos. The dataset comprises 3,839 high-quality video-music pairs, with its data quality ensured by removing vocal tracks and performing manual screening. The development of the dataset involves multi-stage data collection and processing, and it is designed to support research on semantic and rhythmic alignment between videos and music. DVMSet is mainly applied in the field of video background music generation, aiming to solve the problem of automatic alignment between videos and music, and improve the attractiveness of video content and the immersion of audiences.

- 1VidMusician: Video-to-Music Generation with Semantic-Rhythmic Alignment via Hierarchical Visual Features中国科学院自动化研究所 · 2024年



