-- The idea is to detect whether the mushroom is edible or not using Deep Learning. Some of the (al)-most modern approaches are used here.
The main goal is to achieve good Edible Precision due to the fact that it is better not to eat the suspicious mushroom than vice versa.
In the nearest future this project will probably evolve into mushroom classification (#TODO: Parse more mushroom species)
All project is done in Google Colab. The keras framework is used to implement deep learning here. More details in the code.
-- Final score - Edible Precision: 0.97
Possible improvements:
- Parse more mushrooms
- Optimize loss for Edible Precision
- Find anomalies and delete them
- Detech which pictures are mislabeled and add more such pictures/delete them if they are anomalies
- Add more Data Augmentation