MIT builds an AI that detects and creates fake images. Researchers from MIT and IBM have trained an AI to generate photographic images from scratch and intelligently edit objects in them.
The latest tech can benefit artists and designers and offers insight into how neural networks can learn context. The teams hope to enhance the tool to detect fake or altered images as well.
The application is called GANpaint Studio and is available for a free demo. Instead of adding a tree manually to an image, you can tell the toll where you want the object and it will add one that matches the scene. The app also allows you to delete objects from an image, for example, a stool from a kitchen image.
The app is in its development stage, but the research team hopes GANpain Studio will be able to edit video clips one day. They were able to use AI to later add essential props to a film scene they left out initially.
During the development process of the GANpaint Studio, the researcher discovered how the system was able to learn simple rules about relationships between object – like a door does not belong in the sky.
GANpaint Studio uses a GAN, which is a set of neural networks developed to compete against each other. This exposes its internal reasoning for decisions like preventing a cloud from appearing in the grass. These studies can help researchers further understand how neural networks learn context and what we think of as common sense.
GANpaint Studio can not only be able to create fake images, but it can also be able to help computer scientists learn to spot fake images.
Jun-Yan Zhu, who co-authored a paper on the tool said – “You need to know your opponent before you can defend against it,” The researchers will be presenting their work at a conference next month, you can enjoy GANpaint Studies
The researchers will present their work at a conference next month. In the meantime, you can try using GANpaint Studios here.