Proposal Review Link

https://review.udacity.com/#!/reviews/1610899

Capstione Review Link

https://review.udacity.com/#!/reviews/1633836

Drive Link for the restructured dataset

https://drive.google.com/open?id=1Uwtv81b4RQC1nm1o4dti11yabdFW3JK_

Original Datasets and Inputs

The dataset is hosted on Kaggle and is free to download. The data is provided by The Aarhus University Signal Processing group, in collaboration with the University of Southern Denmark
Please visit this link to get the data via Kaggle.

Regarding the Xceptions poor performance

I asked the same question to my mentor on slack and the conclusion was that the Xceptions Bottleneck features wont result in a convergence in case of Logistic Regression as it is. Since I did not wish to change the parameters as it would be unfair to other models in the comparison I let the result be.

Software/Libraries

I used Google Colab and did not require to install any extra libraries

Bottleneck Features and Saved Weights for the benchmark model can be found here

https://drive.google.com/open?id=1wJbzHyD39ADQG_xONv1mfM2q5l5R36jN

For splitting always use the seed as 42 and the ops will obtained
I realized that later that the ground truth should also be stored

The model weights are stored as VGG16_1_1.h5