Researchers from the Max Planck Institute for Informatics, Sarbrücken Research Center, MIT, the University of Pennsylvania and Google have developed an image editing method using GANs (generative adversarial networks) that allows control of almost everything in an image. You put two points and move objects in the photo in three-dimensional space.
“In this work, we explore a powerful but little-explored way to control GANs, namely, dragging any points in an image to precisely reach target points interactively with the user, as shown in Figure 1. To achieve this goal, we propose DragGAN, which consists of two main components: 1) feature-based motion observation that guides the control point to the target position, and 2) a new point tracking approach that uses discriminative GAN features to localize the position of control points.With DragGAN, anyone can change the image with precise control over the movement of pixels, thereby changing the pose, shape, expression and composition of various categories such as animals, cars, people, landscapes, etc."
0 comments
