cGANs with Multi-Hinge Loss
This work addresses the challenge of balancing class conditioning and image quality in GANs for image generation tasks, though it appears incremental as it builds on existing hinge loss methods.
The paper tackles the problem of incorporating class conditional information into GANs by proposing a multi-hinge loss, which improves image quality metrics like Inception Scores and Frechet Inception Distance on the Imagenet dataset.
We propose a new algorithm to incorporate class conditional information into the critic of GANs via a multi-class generalization of the commonly used Hinge loss that is compatible with both supervised and semi-supervised settings. We study the compromise between training a state of the art generator and an accurate classifier simultaneously, and propose a way to use our algorithm to measure the degree to which a generator and critic are class conditional. We show the trade-off between a generator-critic pair respecting class conditioning inputs and generating the highest quality images. With our multi-hinge loss modification we are able to improve Inception Scores and Frechet Inception Distance on the Imagenet dataset. We make our tensorflow code available at https://github.com/ilyakava/gan.