EmoGan: Generating Realistic Emotional Pictures with Matched Visual and Emotional Attributes
Stanford University
In collaboration with Jinxiao Zhang
In collaboration with Jinxiao Zhang
2021
In the field of psychology, researchers need to make a specific claim about emotional responses that correspond only to emotional representations while controlling for the low-level visual features across different emotional categories. We created an EmoGAN that generates realitic emotional images while preserving visual and emotional attributes.
Architecture of EmoGAN
GAN model arcitecture. This model consists of an U-Net generator with skipped connections and a two discriminators. One discriminator matched pixel-wise image information while the other discriminator matched emotional sentiment of the image.EmoGAN generates more realistic pictures than the conventional low-level matching model
