Computer Vision

With a new generation of hardware (AI chips, stronger GPUs), we see a lot of advances in image processing. Generative Adversarial Networks (GANs) since their introduction in 2015 became widespread and continue to surprise with their results. In particular, the next frontiers in image generation include commercial uses of GANs in the fashion and film industry; some we have discussed in the previous chapter.

In fashion, a standard process is organising a photo session for each new line of products. One needs high-quality photos with constraints on what’s shown, namely a garment from a fashion designer. GANs could potentially be used to generate models on which fashion products are shown. Some experiments in this direction are already in place, but the whole problem - generating photorealistic images of models with particular pieces of clothes on them - seems still to be out of reach. Its solution will require both local and global GANs for managing the generation...