I did have to do some amount of flickr scraping from the dataset. The browser also aligns and crops the faces according the way nvidia did as well. I thought it would be an interesting commentary to look at the faces in the same way that the machine did.