Instrument Classfication Challenge for the OEAW AI Summer School: https://www.kaggle.com/c/oeawai/
Our score in private board was 0.77231 (F1). It was achieved by an geometric average of the outputs of two models:
SpectralResNet18 (private score alone: 0.75906) learns the MFC trained for 9 epochs with Adam Optimizer, lr 0.0001, batch size 8 on GeForce GTX1080
MSResNet (private score alone: 0.71199) learns the raw audio trained for 10 epochs with Adam Optimizer, lr 0.0001, batch size 8 on GeForce GTX1080