This release contains preregistered split definitions, exact source identities, acquisition records, verification hashes, generation revision records, and compact reference evaluation outputs for the osu-ai-detector technical report.
It intentionally excludes third-party raw beatmaps, audio, and upstream model weights. Researchers acquire those files from their original sources by using the published scripts, and the scripts reject… See the full description on the dataset page:
https://huggingface.co/datasets/NettoAndTetto/osu-ai-detector-reproducibility.
