BirdCLEF 2022

Introduction

This repository contains source code generated by Luminide. It may be used to train, validate and tune deep learning models for audio classification. The following directory structure is assumed:

├── code (source code)
├── input (dataset)
└── output (working directory)

The dataset should have audio files inside a directory named train_audio and a CSV file named train_metadata.csv. An example is shown below:

input
├── train_metadata.csv
└── train_audio
    ├── 800113bb65efe69e.ogg
    ├── 8002cb321f8bfcdf.ogg
    ├── 80070f7fb5e2ccaa.ogg

The CSV file is expected to have labels under a column named primary_label as in the example below:

filename,primary_label
800113bb65efe69e.ogg,pacific_wren
8002cb321f8bfcdf.ogg,hermit_thrush pacific_wren
80070f7fb5e2ccaa.ogg,hermit_thrush

If an item has multiple labels, they should be separated by a space character as shown.

To use this repo with Luminide

• Accept competition rules.
• Attach a Compute Server that has a GPU (e.g. gcp-t4).
• Configure your Kaggle API token on the Import Data tab.
• On the Import Data tab, choose Kaggle Competition Data and then enter birdclef-2022.
• Install required packages on the Compute Server using the Run Experiment menu (select Custom > install.sh).
• Train a model using the Run Experiment menu.

Kaggle submission

• Upload the code to Kaggle as a dataset by using the Run Experiment menu (select Custom > kaggle.sh).
• To create a submission, copy kaggle.ipynb to a new Kaggle notebook.
• Add the notebook output of https://www.kaggle.com/luminide/wheels1 as Data.
• Add your dataset at https://www.kaggle.com/<kaggle_username>/kagglecode as Data.
• Add the relevant competition dataset as Data.
• Save the notebook after turning off the Internet setting and turning on the GPU.
• Submit the results and wait for the notebook to finish.
• Check the leaderboard to see your score!

Additional features

• Use the Experiment Tracking menu to track experiments.
• To tune the hyperparameters, edit sweep.yaml as desired and launch a sweep from the Run Experiment tab. Tuned values will be copied back to a file called config-tuned.yaml along with visualizations in sweep-results.html.
• To use the tuned hyperparameter values, copy them over to config.yaml before training a model.
• For exploratory analysis, run eda.ipynb.
• To monitor training progress, use the Experiment Visualization menu.
• After an experiment is complete, use the file browser on the IDE interface to access the results on the IDE Server.
• To generate a report on the most recent training session, run report.sh from the Run Experiment tab. Make sure Track Experiment is checked. The results will be copied back to a file called report.html.

Acknowledgements

• The network architecture for the attention head is adapted from a notebook by tattaka

For more details on usage, see Luminide documentation