This repository contains scripts which can be run from the command line and utilize mir_eval to compute metrics according to reference and estimated annotations you provide.
To use the evaluators, you must first install mir_eval and its dependencies (see here).
By way of example, we'll cover the usage of the beat detection evaluator beat_eval.
To use an evaluator for a different task, simply replace beat_eval in the following with the name of the evaluator for the task you're interested in.
To get usage help, simply run
./beat_eval.py --help
As an example, to evaluate generated beat times stored in the file estimated_beats.txt against ground-truth beats stored in the file reference_beats.txt and store the resulting scores in results.json, simply run
./beat_eval.py -o results.json reference_beats.txt estimated_beats.txt
The file results.json will now contain the achieved scores in machine-parsable, human-readable json format. Nice!