Science Fair code
-
Only tested on python 3.6.4
-
Uses keras with tensorflow, takes at least an hour per training pass.
-
Each file runs the specific model five times with random starting weights.
-
The CNN model is currently do terrible compared to the results seen in all my research, as well as the MLP recieving really high results.
-
Setup a default, cookie cutter Virtual Environment
-
Download the Extended Yale Database B from -> http://academictorrents.com/details/06e479f338b56fa5948c40287b66f68236a14612
-
Put the images from the database in a folder named ExtendedYaleB at the same level as the model python files.
-
Run the Convolutional Neural Network and Multi-Layer Perceptron. These will most likely create the text file, which shows the duration of training and the achieved accuracy.