Single class image classification using deep learning

Question

I am familiar with binary and multi classification problems using CNNs but not sure how to build a classifier to classify an image as 'object' or 'not object' without overfitting when my train data has a single label (i.e. only images that contain 'object').

What I want to do is similar to what was done here but I am only allowed to use CNN for training and cannot add other images to the dataset.