Artificial Intelligence is increasingly visible on what it can achieve in this new era of computations and massive data storage. Until recently there was not enough computational power to perform anything miraculous by the AI algorithms, most of which have been defined in research papers 30-50 years old. These techniques ain’t new. Their implementations and effects are visible now.
I have been discussing with my friend around possibility to create neural networks to perceive images and predict functions to compute a procedural generator from it. Below is what another friend sent accidentally and is worth sharing. This is a research paper and a video around how AI can perceive images and build 3d models from it.
You must already be aware about the projects like Deep Style Transfer which can transfer image styles and apply to completely different photo. Look at the end of link to see the images and their effects. This is the kind of visual power AI is coming up with. Adobe has recently developed this technology with Cornell Researchers behind this innovative effort in attempts to safeguard the future for themselves. [Reference]
Than there is CycleGAN which can understand the maps between two images and use it to create similar images. It can make horses to zebras and vice versa, apply seasons on images you can not imagine and mimic artistic styles. This is one of the top github trending projects now.