cGANs with Projection Discriminator
This work addresses the challenge of conditional image generation for computer vision applications, offering a novel architectural improvement over existing methods.
The authors tackled the problem of conditional GANs by proposing a projection-based discriminator that incorporates conditional information more effectively than standard concatenation methods, resulting in significant improvements in class-conditional image generation on the ImageNet dataset and enabling high-quality super-resolution and category transformation.
We propose a novel, projection based way to incorporate the conditional information into the discriminator of GANs that respects the role of the conditional information in the underlining probabilistic model. This approach is in contrast with most frameworks of conditional GANs used in application today, which use the conditional information by concatenating the (embedded) conditional vector to the feature vectors. With this modification, we were able to significantly improve the quality of the class conditional image generation on ILSVRC2012 (ImageNet) 1000-class image dataset from the current state-of-the-art result, and we achieved this with a single pair of a discriminator and a generator. We were also able to extend the application to super-resolution and succeeded in producing highly discriminative super-resolution images. This new structure also enabled high quality category transformation based on parametric functional transformation of conditional batch normalization layers in the generator.