ilyassmoummad / Mix2

Mix2 (Mixture of Mixups), a framework to handle multi-label and class imbalance. Experiments on AnuraSet, a dataset of anuran sounds. (EUSIPCO 2024)

Mixture of Mixups for Multi-label Classification of Rare Anuran Sounds

Authors: Ilyass Moummad, Nicolas Farrugia, Romain Serizel, Jeremy Froidevaux, Vincent Lostanlen

Preprint: https://arxiv.org/abs/2403.09598

---

Update : This work is accepted for EUSIPCO 2024 (Special Session: Signal Analysis for Biodiversity)

We introduce a framework that leverages mixing regularization methods Mixup, Manifold Mixup, and MultiMix to handle multi-label and class imbalance on the Anuraset dataset.

We base our code on the official implementation of AnuraSet baseline: https://github.com/soundclim/anuraset where you can find the link to download the dataset AnuraSet
Python libraries required : torch, torchmetrics, numpy, pandas, tqdm

main.py: main code for training and evaluating on AnuraSet
dataset.py: dataset class
models.py: code for MobileNetV3 model
train.py: train utility functions
val.py: evaluation utility functions
transforms.py: transformation classes
args.py: argparse of the arguments

Training and Evaluation

python3 main.py --rootdir dataset_path --mix mix2 --device 'cuda' --sr 16000 --workers 16 --save

To cite this work:

@misc{2403.09598,
Author = {Ilyass Moummad and Nicolas Farrugia and Romain Serizel and Jeremy Froidevaux and Vincent Lostanlen},
Title = {Mixture of Mixups for Multi-label Classification of Rare Anuran Sounds},
Year = {2024},
Eprint = {arXiv:2403.09598},
}