This repository represents Dieters part of teams NVBird solution to the BirdCLEF24 competition on kaggle.
I used the nvcr.io/nvidia/pytorch:24.03-py3 container from the ngc catalog to have a consistent environment between team members. You can run it via
docker run --gpus all -it --rm nvcr.io/nvidia/pytorch:24.03-py3
Within the container clone this repository and install necessary packages with
git clone https://github.com/ChristofHenkel/kaggle-birdclef24-3rd-place-solution-dieter
cd kaggle-birdclef24-3rd-place-solution-dieter
pip install -r requirements.txt
By default training is logged via neptune.ai. You need to use your own neptune project which is set via cfg.neptune_project in configs/cfg_XXX.py
We have preprocessed the dataset into numpy format files for fast loading.
After downlaoding the competition data to /mount/birdclef24/data/ run scripts/convert_np_first_last_10_v2.py to create npy files containing wav signal of first and last 10sec of competition data.
Run scripts/create_train_folded_v3c.py to process the original comeptition meta data and handle some duplicates and create folds.
Run scripts/create_test_fake_gt.py to enable prediction on unlabled soundscapes.
Download model weights from AVES and put torchaudio config and weights of birdaves-biox-base into /mount/birdclef24/data/
Download background noise from https://www.kaggle.com/datasets/honglihang/background-noise and put into /mount/birdclef24/data/background/
run the following commans to train 5 seeds each for 3 different model configs
5x python train.py -C cfg_1 --fold -1
5x python train.py -C cfg_2 --fold -1
5x python train.py -C cfg_3 --fold -1
This results in 15 csv files pl_df_SEED in respective model folders. Simply average predictions from 3 models each to create 5x different
pl_blended_0.csv
pl_blended_1.csv
pl_blended_2.csv
pl_blended_3.csv
pl_blended_4.csv
needed for round 2
run
python train.py -C cfg_pl_1 --fold -1 --pl_df pl_blended_0.csv
python train.py -C cfg_pl_1 --fold -1 --pl_df pl_blended_1.csv
python train.py -C cfg_pl_1 --fold -1 --pl_df pl_blended_2.csv
python train.py -C cfg_pl_1 --fold -1 --pl_df pl_blended_3.csv
python train.py -C cfg_pl_1 --fold -1 --pl_df pl_blended_4.csv
to get final model weights
see public notebook