README.md

Image2Music

Article: Zijian Zhao*, Dian Jin, Zijing Zhou, "Zero-Effort Image-to-Music Generation: An Interpretable RAG-based VLM Approach", ACM ICMR 2026

1. Worflow

2. Dataset

amaai-lab/MidiCaps · Datasets at Hugging Face

Please rename the train.json as meta.txt.

The data process part is based on the code of jwdj/EasyABC: EasyABC (github.com).

3. Run the Model

python main.py

4. Citation

@article{zhao2025zero,
  title={Zero-Effort Image-to-Music Generation: An Interpretable RAG-based VLM Approach},
  author={Zhao, Zijian and Jin, Dian and Zhou, Zijing},
  journal={arXiv preprint arXiv:2509.22378},
  year={2025}
}

5. Links

Some websites provide the service for abc2midi and midi2abc:

midi2abc (marmooo.github.io)

ABC notation converter - Nota ABC (notabc.app)