Downloads

Run your model on standard video and score its event predictions.

The 40 public recordings are for development. Published results use separate evaluation sets.

Get started

1. Download and extract the code and videos into the same directory.
2. Open the onsetbench directory and install the evaluator.
3. Add a model adapter and run the benchmark.

cd onsetbench
python -m pip install .
python -m onsetbench run --dataset data/reassemble_public40.json --inputs ../inputs/inputs.json --predictor my_adapter:predict --name "My model" --out my-run

The runner saves predictions and scores to my-run.