Change default parameters for tilenet model and add training augmentation

dsgt-birdclef / birdclef-2022

Code for the BirdCLEF 2022 competition by the DS@GT team

2 stars 3 forks source link

Change default parameters for tilenet model and add training augmentation #30

Closed acmiyaguchi closed 2 years ago

acmiyaguchi commented 2 years ago

This adds the original parameterization for the tilenet model (with 20m parameters instead of 1m parameters). This should hopefully allow for better learning. In addition, this PR will also include data augmentation in the dataloader.