In step 1, we simply clone the repository for the deep steganography project. Some background on the theory and implementation of this project can be found in the paper Hiding Images in Plain Sight: Deep Steganography (https://papers.nips.cc/paper/6802-hiding-images-in-plain-sight-deep-steganography).
The basic idea is that there is a hiding network (H-net) and a reveal network (R-net), both of which are trained adversarially. Continuing to step 2, we prepare our pretrained model. The model that we used here was trained on 45,000 images from ImageNet, and evaluated on 5,000 images. All of the images were resized to 256 × 256 without normalization and the task took 24 hours of training on a single NVIDIA GTX 1080 Ti. Next, we ...
