The aim of this project is to categorise animals from images provided in iWildCam 2019 - FGVC6 Kaggle dataset.
In this priject I used transfer learning.
I'm comparing 3 approaches, where all images resized to 100x100
- with some augmentation, all images used
- more augmentation and calculated class weights to reduce impact of imbalanced data
- only 20% of
emptyimages used
TODO
- Try different models
- Two step iteration with image segmentation