From: Training audio transformers for cover song identification
Results
Methods
MAP
MR1
Standard Triplet Loss(F0) [12]
0.222
-
Standard Triplet Loss(Multi-F0) [12]
0.280
RE-MOVE [39]
0.457
1388
ASimT (\(\mathcal {L}_{CE}+\mathcal {L}_{CON}\))
0.301
1234
ASimT (\(\mathcal {L}_{CE}+\mathcal {L}_{MAP}\))
0.466
1048