Added in some k-fold cross-validation for training the last few layers of VGG-16, finished the assignment.

Added code for knn.py which implements KNN on the output of VGG.

Fixed the issue we were having with predicting elements of the test set. We deprecated our training set so that we were using just the orginal training set. For some reason we had to use the imagegenerator from keras in order to get good corrections.

Merge branch 'master' of https://github.com/keenan-burnett/SceneRecognition

Minor Change.

Changed everything over to use only the data that was given to us

added the rest of my changes.

Made modifications to my code so that it will properly print our predictios to a CSV file, we also now grab all the text examples from the two different folders.

Finished modifying keras_load... so that we can load the model and run it on the test set. It will also output the results to a csv file.

The imageNet1 model is now working and has early stopping implemented. It takes about 14 epoches to train to ~72% accuracy on the validation data

Merge branch 'master' of https://github.com/keenan-burnett/SceneRecognition

getting the imageNet1 model to work, I think this involved changing a directory and making sure all of the images were sized and counted correctly

Working on adding some features to the code like being able to load weight from both VGG and FC.

Minor fix to loading labels.

Made some changes to the keras python files, notably that we no longer shuffle the training example being fed into VGG16 as this doesn't need to be randomized when we're just trying to get the output of the net for later use in training the fully connected layers. Upsized the input image to 224x224.