<p>In creating visual content that meets user needs, flexible and precise control over pose, shape, expression, and layout of generated objects is often required. This work explores a more efficient and less studied way of controlling GANs, allowing users to 'drag' any point in an image to accurately reach target points in an interactive manner. For this, we propose DragGAN, which consists of two main components: 1) feature-based motion control that directs the control point to the target position, and 2) a new point tracking approach that uses discriminative features of GANs to determine the position of control points.</p>
<p>With DragGAN, anyone can deform an image with precise control over pixel movement, thereby altering the pose, shape, expression, and layout of various categories such as animals, cars, people, landscapes, etc. Since these manipulations are performed on a trained generative image from a GAN, they can produce realistic results even in complex scenarios, such as recreating occluded content and deforming shapes that consistently follow the rigidity of the object.</p>
<p>🔗 Website: <a href="https://vcai.mpi-inf.mpg.de/projects/DragGAN/">https://vcai.mpi-inf.mpg.de/projects/DragGAN/</a><br />📝 Paper: <a href="https://vcai.mpi-inf.mpg.de/projects/DragGAN/data/paper.pdf">https://vcai.mpi-inf.mpg.de/projects/DragGAN/data/paper.pdf</a><br />🧪 Demo: <a href="https://huggingface.co/spaces/radames/DragGan">https://huggingface.co/spaces/radames/DragGan</a><br />👨💻 Github: <a href="https://github.com/XingangPan/DragGAN">https://github.com/XingangPan/DragGAN</a></p>;
<p>#ai #gan #image</p>