For the both giants, namely Facebook and Google, it seems that recognizing an image is not enough for them, but also, they want to create the images too. The both tech giants have tried to show off the neural networks, which automatically creates or generate images on the basis of what an object looks like.
The social media giant approach uses two of the following techniques to generate tiny thumbnail images automatically. The technique used by the Facebook is somewhat similar to the random vector, and it is similar to how people used to learn painting from the scratch.
The primary algorithm creates a picture, which is purely based on the random vector and creates images using them. The secondary algorithm used to check whether the images look realistic or not. If the image generated by the first one is not realistic, then it will be rejected automatically. As a result of both, the user will get more appropriate as well as convincing results.
The research states,
“Around 40 percent of the samples generated by our class conditional LAPGAN model are realistic enough to fool a human into thinking they are real images,”
The search engine giant takes the approach in the right opposite manner, instead of looking for the realism, the algorithm produces art by allowing the neural networks to run wild and automatically decide on the visual elements that it want to create. If an input is passed to the machine, the photo will be analyzed and refined the images until “they’re impossible to miss”.
For instance, an image of the sky is provided to the machine, the machine will think, there are birds in the background. The image will be automatically refined until “they’re impossible to miss”. The skills by the programs created by Google and Facebook aren’t replacing human artists, but they possess the skills to draw images automatically.