Reproduce paper results¶

You may follow the following steps to reproduce the results presented in the ACMMM2016 paper “Event Localization in Music Auto-tagging”

Datasets¶

Training:

MagnaTagATune

Frame-level evaluation:

MedleyDB

Feature extraction¶

scripts/extract_feats.py

Make features and annotations into experiment data¶

???

Standardize experiment data¶

scripts/standardize_data.py

Train model¶

scripts/train_model.py

Get thresholds¶

With MagnaTagATune data:

get_thresholds.with_magnatagatune.py

With MedleyDB data:

get_thresholds.with_medleydb.py

Evaluate model¶

For clip-level prediction:

test_model.clip.py

For frame-level prediction:

test_model.frame.py