Video-Audio CAptions Dataset (VACAD)
收藏资源简介:
MusiScene数据集由人工智能大学的研究团队创建,旨在探索音乐语言模型在场景想象方面的能力。该数据集包含3371对视频和音频配对,以及相应的视频和音乐描述。研究人员利用现有的MU-LLaMA模型进行微调,以便使其能够回答与场景相关的问题,例如“这段音乐适合哪种视频?”研究团队希望利用生成的场景想象描述来提高视频背景音乐生成的质量。
The MusiScene dataset was developed by a research team from the University of Artificial Intelligence, with the aim of exploring the scene imagination capabilities of music language models. This dataset contains 3371 pairs of matched video and audio clips, along with corresponding descriptions of the videos and music. Researchers fine-tuned the existing MU-LLaMA model to enable it to answer scene-related questions, such as "What type of video is this music suitable for?". The research team intends to utilize the generated scene imagination descriptions to enhance the quality of video background music generation.




