Music4way-MI2T, Music4way-MV2T, Music4way-Any2T
收藏资源简介:
Music4way-MI2T、Music4way-MV2T和Music4way-Any2T是三个用于音乐理解任务的多模态数据集,由索尼集团株式会社创建。这些数据集将音乐、文本、图像和视频四种模态进行对齐,旨在提升多模态音乐理解模型的能力。Music4way-MI2T和Music4way-MV2T分别包含音乐与图像/视频以及文本的配对,而Music4way-Any2T则提供了更灵活的模态组合,以评估模型对不同输入和查询的鲁棒性。这些数据集的构建基于AudioSet音乐剪辑,并通过多种特征提取方法生成了丰富的文本描述,以促进音乐、图像、视频和文本的深度融合。
Music4way-MI2T, Music4way-MV2T, and Music4way-Any2T are three multimodal datasets for music understanding tasks, developed by Sony Group Corporation. These datasets align four modalities—music, text, image, and video—with the goal of enhancing the capabilities of multimodal music understanding models. Music4way-MI2T and Music4way-MV2T respectively contain paired samples associating music with images or videos and their corresponding text, while Music4way-Any2T offers more flexible modality combinations to evaluate the robustness of models against diverse inputs and queries. These datasets are built upon AudioSet music clips, and rich textual descriptions are generated through multiple feature extraction approaches to promote the deep fusion of music, images, videos and text.

- 1DeepResonance: Enhancing Multimodal Music Understanding via Music-centric Multi-way Instruction Tuning索尼集团株式会社 · 2025年



