The threshold is the probability of an image to be a tobacco advertisement and whether we classifiy it as such. The initial data we have for training is from a previous NYU CUSP student research group, which we refer to as MD17. In total, there are 300 different images with tobacco advertisements. 300 images were augmented to generate 9,331 total images by manipulating image characteristics such as color balance, contrast levels, and random noise. Different attempts of selecting training set and validation set were made.  For example, a random sample of 70% of the 9,331 images as training and 30% as validation has been deemed inappropriate since there was considerably high overlap between training and validation set after image augmentation . This was established with cosine similarity and hashing  by drawing parameters from a random distribution. Considering the limited initial data we had, choosing a pre-trained model is crucial, and is also a typical research approach which takes advantages of previously established work.  ResNet101 (He 2016), a  neural network with 101 residual layers, was used.