The data set if correctly organised. Might need to experiment with lower batch size or change the model.

Don't have labeled data for FDDB to do object localisation, therefore I need to change the model so that I do cross-validation only on the regression layer.

Redid the numpy arrays. Need to change the model to take a normal numpy array as input (super inefficient).

Installed Keras alog with all dependencies and updated the requirements.txt file with packages needed only from now on.

Cleaned the repo, installed miniconda. Successfully created environment for tensorflow with GPU support. Tested in NVIDIA TITAN X

Redone run 1 of FDDB. The performance results are now consistent. They need to be analysed to see if there is any obvious relationship between the any combination of parameters and the performance of Facenet

Ran condor job. There were a lot of unexpected exceptions thrown out. Fixed one file whose id stamp was not changing