Ein bedingtes Generatives Gegennetzwerk (Conditional GAN or cGAN) is an advanced variant of the traditional Generative Adversarial Network (GAN) that allows for the generation of data conditioned on specific inputs or labels. In a standard GAN, two neural networks—the generator and the discriminator—compete against each other: the generator creates fake data, while the discriminator attempts to distinguish real data from the generated data. The cGAN enhances this framework by introducing additional information, which guides the generator in producing data that aligns with the specified condition.
In a cGAN, both the generator and the discriminator receive additional input in the form of labels or conditions. For instance, if the task is to generate images of handwritten digits, the generator would be conditioned on the desired digit label (0-9). This conditioning enables the generator to create images that not only appear realistic but also correspond to the specified digit. On the other hand, the discriminator also uses this label information to assess whether a generated image ist eine realistische Darstellung der gegebenen Ziffer.
Das architecture of a cGAN typically employs a similar setup to that of a standard GAN, but with modifications to incorporate the conditioning variables. This approach broadens the applicability of GANs, making them suitable for tasks such as Bild-zu-Bild-Übersetzung, text-to-image synthesis, and even generating audio or video based on specific attributes. Overall, Conditional GANs represent a powerful tool in the field of generativen Modellen, enabling more controlled and meaningful data generation.