This repository contains the implementation of our (team DBIS) solution to the AcousticBrainz Genre Task: Content-based music genre recognition from multiple sources as part of MediaEval 2017.
Our solutions to subtasks 1 and 2 can be found in the task1 and task2 folders, respectively. The main code files are task1.py and task2.py.
The detailed list of features we used to train our classifiers is given in the file features.txt.
Pickled datasets for subtask 1 and pre-trained classifiers for subtask 2 can be downloaded at the following locations:
- Datasets for subtask 1: https://dbis-owncloud.uibk.ac.at/index.php/s/PL7iQzaZBGpzWhm/download?path=%2F&files=task1_datasets.zip
- Classifiers for subtask 2: https://dbis-owncloud.uibk.ac.at/index.php/s/PL7iQzaZBGpzWhm/download?path=%2F&files=task2_classifiers.zip
Example usage for task1.py:
./task1.py -i discogs.pickle -o out.txt
Example usage for task2.py:
./task2.py -c1 classifiers/discogs.pickle -c2 classifiers/allmusic.pickle -c3 classifiers/lastfm.pickle -c4 classifiers/tagtraum.pickle -n1 discogs -n2 allmusic -n3 lastfm -n4 tagtraum -m genre_mapping.csv -test data/discogs.pickle -o out.txt