EC-GAN: Emotion-Controllable GAN for Face Image Completion

Image completion methods based on deep learning, such as Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs), have succeeded in producing semantically plausible results. However, existing facial image can either produce only one result or, although they provide multiple results, cannot attribute particular emotions to the We propose EC-GAN, a novel Emotion-Controllable GAN-based model that infer customize generative emotions. an emotion inference module infers of faces unmasked regions faces. The is trained supervised manner enforces encoder disentangle semantics from native latent space. also developed control modify codes emotions, moving initial toward desired while maintaining remaining features. Extensive experiments were conducted two datasets, CelebA-HQ CFEED. Quantitative qualitative results indicate EC-GAN produces images with diverse expressions even when main features are masked. On other hand, promotes semantic capability irregularly masked holes, resulting more natural expressions.