blobgan
BlobGAN: A BIG step for GANs
I explain Artificial Intelligence terms and news to non-experts. BlobGAN allows for unreal manipulation of images, made super easily controlling simple blobs. All these small blobs represent an object, and you can move them around or make them bigger, smaller, or even remove them, and it will have the same effect on the object it represents in the image. As the authors shared in their results, you can even create novel images by duplicating blobs, creating unseen images in the dataset like a room with two ceiling fans! Correct me if I'm wrong, but I believe it is one of, if not the first, paper to make the modification of images as simple as moving blobs around and allowing for edits that were unseen in the training dataset.
Editing a GAN's Latent Space With 'Blobs'
New research from UC Berkeley and Adobe offers a way to directly edit the hyperreal content that can be created by a Generative Adversarial Network (GAN), but which can't usually be controlled, animated, or freely manipulated in a manner long familiar to Photoshop users and CGI practitioners. Titled BlobGAN, the method involves creating a grid of'blobs' – mathematical constructs that map directly to content within the latent space of the GAN. By moving the blobs, you can move the'objects' in a scene representation, in an intuitive manner that's nearer to CGI and CAD methods than many of the current attempts to map and control the GAN's latent space: Scene manipulation with BlobGAN: as the'blobs' are moved by the user, the disposition of latent objects and styles in the GAN are correspondingly altered. For more examples, see the paper's accompanying video, embedded at the end of this article, or at https://www.youtube.com/watch?v KpUv82VsU5k Since blobs correspond to'objects' in the scene mapped out in the GAN's latent space, all the objects are disentangled a priori, making it possible to alter them individually: Objects can be resized, shrunk, cloned, and removed, among other operations. Blobs can be duplicated in the interface, and their corresponding latent representations will also be'copied and pasted'.