Remember our good old friend, the GAN? It’s the type of AI algorithm that synthesizes art and realistic images by pitting two neural networks against each other. Here’s another thing it can do: remove photo-bombers from your pictures, sans other photo-editing software.
Researchers at the Max Planck Institute and CISPA Helmholtz Center developed a technique that uses two separate GANs to achieve the task. The first GAN uses what they call a “mask generator” to clumsily cover up the pixels of the target object until the “mask discriminator” can no longer recognize it. The second GAN uses an “in-painter” to turn the messy cover-up into something photorealistic, until the discriminator can no longer tell that it’s synthesized. The result? See for yourself above.
- Rakshith Shetty et al.