Conditional Image Generation with One-Vs-All Classifier
This work addresses conditional image generation for computer vision applications, presenting an incremental improvement over existing conditional GANs.
The paper tackles conditional image generation by proposing GAN-OVA, which replaces the standard discriminator with a One-Vs-All classifier to improve training stability and generation quality. Experimental results on MNIST and CelebA-HQ datasets show progress in stable training, faster generation across classes, and enhanced quality.
This paper explores conditional image generation with a One-Vs-All classifier based on the Generative Adversarial Networks (GANs). Instead of the real/fake discriminator used in vanilla GANs, we propose to extend the discriminator to a One-Vs-All classifier (GAN-OVA) that can distinguish each input data to its category label. Specifically, we feed certain additional information as conditions to the generator and take the discriminator as a One-Vs-All classifier to identify each conditional category. Our model can be applied to different divergence or distances used to define the objective function, such as Jensen-Shannon divergence and Earth-Mover (or called Wasserstein-1) distance. We evaluate GAN-OVAs on MNIST and CelebA-HQ datasets, and the experimental results show that GAN-OVAs make progress toward stable training over regular conditional GANs. Furthermore, GAN-OVAs effectively accelerate the generation process of different classes and improves generation quality.