# 101st Place Solution (Top 8%)

This competition was a toughie. Special thanks to my team members for their support
[@doanquanvietnamca](https://www.kaggle.com/doanquanvietnamca)
,
[@truonghoang](https://www.kaggle.com/truonghoang)
throughout this competition.
As this was my first audio competition, in the beginning, I tried effnetb3 using fastai as a baseline, with precomputed spectrograms. This was able to beat the public LB baseline, giving a score of 0.554. However, I was unable to improve this model. My team and I subsequently tried several other new models in the process
Models tried:
EfficientNetB1
ResNest50 (best model)
MobileNetV2
SED model, with effnetb0 backbone
EfficientNetB3 (with pretrained noisystudent weights)
ResNest50 performed well for this task, with the same classifier head provided by
[@hidehisaarai1213](https://www.kaggle.com/hidehisaarai1213)
.
Augmentations also worked well for us, namely:
Adding gaussian noise
Adding pink noise
Adding background noise
We also tried a model with 265 classes, with nocall as a class but that didn't work better than a normal 264 classifier model.
We tried incorporating secondary labels to perform the multilabel classification task, it gave decent CV (up to 0.7x validation micro score), faltered in public LB (0.569) but was the best performing model for private LB (0.616).
We were unable to bridge the gap between CV and LB, the correlation between CV and LB was not strong so in the end we had to trust our methodology and public LB.
We settled for an ensemble of models between ResNest50 and MobileNetV2 which landed us with the bronze solution.
Summary of things that did not work for us:
SED ( could not get it to work somehow, will have to learn from training solution by
[@hidehisaarai1213](https://www.kaggle.com/hidehisaarai1213)
to understand more why)
265 class classifier (including nocall class), decreased public and private LB
Denoising on test set
F1Loss (possibly due to batch size being too small)
FocalLoss
Special thanks to the organizers of this competition and fellow competitors whom I learn loads from (@hidehisaarai1213 ,
[@radek1](https://www.kaggle.com/radek1)
,
[@kneroma](https://www.kaggle.com/kneroma)
,
[@ttahara](https://www.kaggle.com/ttahara)
). Hope to join another audio competition to further my learning. Congrats to all winners in this competition.