Contrastive Pre-Training for BirdCLEF Classification

This is a simple baseline for pre-training of BirdCLEF classification. The official website of this competetion is on this kaggle page. We used publically available BirdCLEF 2021 and BirdCLEF 2022 dataset for contrastive pretraining using SimSiam.

Train

torchrun --nproc_per_node=4 main.py
    --mode train
    --config [path to config file]
    --workdir [name of work directory]

Finetune

torchrun --nproc_per_node=4 main.py
    --mode tune
    --config [path to config file]
    --workdir [name of work directory]
    --tunedir [name of finetune directory]

Evaluation

Evaluation codes are ignored due to kaggle competetion.