kaggle-birdcall-resnet-baseline-training

The repository to show the way I trained https://www.kaggle.com/hidehisaarai1213/birdcall-resnet50-init-weights

Prerequisites

• Python >= 3.6
• ffmpeg
• sox
• GPU (P100 / V100) is the best

Intstallation

pip install -r requirements.txt

To reproduce

1. First you need to put dataset in /input/birdsong-recognition. The data structure should be the same as that of Kaggle dataset.
2. Run make prepare at the top of this repository. This command will perform resampling the dataset to 32kHz. Note that this will take a bit long (1 Hours on NVIDIA DGX Station).
3. Run make train at the top of this repository.