search

bdzyubak / tensorflow-sandbox

A repository for studying applications of Deep Learning across fields, and demonstrating samples of my code and project managment
0 stars 0 forks source link

Implement EfficientNet for pathology slide classification #11

Closed bdzyubak closed 2 years ago

bdzyubak commented 2 years ago

Implement the EfficientNet architecture to run for a classification task. The first example will be classification of pathology tumor slides from Kaggle. https://www.kaggle.com/code/taylorkern/tk-efficientnet

The generalized model initializer (https://github.com/bdzyubak/Bogdans-Deep-Learning-Sandbox/issues/9) and data preparer (https://github.com/bdzyubak/Bogdans-Deep-Learning-Sandbox/issues/10) will need to be extended/tested for classification tasks.

bdzyubak commented 2 years ago

Implemented in 6fd7f4f8b50ad9a5038a93119a84b60cecb2017b. EfficientNET works on the KaggleDataset yielding 93% accuracy at this time. It is supported by the generalized model initializer and data fetcher. Local minima bypasser yielded an improvement from 89% original accuracy on the third iteration! https://github.com/bdzyubak/Bogdans-Deep-Learning-Sandbox/issues/12