The following project placed 2nd at the annual 24 hour UMW Hackathon 2021 [code formatting is reflective of time constraints]
Collected the data and trained the model for the following models;
- Eagle: NonEagle(other birds) classifier -Reached a reported accuracy of roughly 91%
- Head CT Scan(control:necrosis:tumor) classifier -Reached a reported accuracy of 32% using a data set of roughly 80 images curated from the internet -Includes a data upload and relabeling functionality that then generates a new model for continued model training
The Goal Of The Project:
The Eagle classifier is a proof of concept of how good a classifier can be generated from a personal data set and trainined over just a few layers
The Head CT Scan notebook is meant to function as a tool for radiologist and researchers who have a limited amount of data and will be manually labeling their data as they progress through their research.
Main libraries: Tensorflow Keras Numpy fastai pickle
References:
1)https://www.youtube.com/watch?v=j-3vuBynnOE&list=LL&index=3 2)https://www.youtube.com/watch?v=WvoLTXIjBYU&list=LL&index=4&t=937s 3)https://www.youtube.com/watch?v=iGWbqhdjf2s 4)https://keras.io/examples/vision/image_classification_from_scratch/