MUVideo: 13,203 Music Files for Image-to-Music Generation

13,203 music files with a total playtime of 36.72 hours, generated using the MU-LLaMA and VideoMAE captioning models. The dataset was created by M2UGen to train the M2UGen model and was last updated on 2024-01-02.

Use Cases

• Training image-to-music generation models based on the paired video and audio files.
• Researching cross-modal alignment between visual content and generated music.
• Benchmarking generative audio models using the structured MUVideoInstructions.json file.
• Fine-tuning captioning models for video-to-text tasks based on the described conversation format.

Strengths

• Contains 13,203 distinct music files.
• Provides a substantial 36.72 hours of total audio content.
• Includes structured instruction data (MUVideoInstructions.json) linking video inputs to audio outputs.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the instruction file is unknown, which may limit suitability assessment.
• Last updated 2024-01-02 08:05:53; freshness should be verified for current research.

Provenance

Source

M2UGen

Collection Method

Generated using MU-LLaMA and VideoMAE captioning models.

Time Range

null

Freshness

Last updated 2024-01-02 08:05:53.

Geography

null