Mixed Feelings

Recognizing, representing, and interpolating human emotions with deep learning.

Using learned representations of emotion to project new emotions onto a neutral face.

Myself, Daniel Jeong, and Anush Devadhasan (my PhD colleagues) a VAE-based architecture for learning latent representations of human emotions. Check out our in-depth presentation here!

We disentangled learned emotion representations by applying an adversarial discriminator to the latent space. Then we used these disentangled representations to interpolate between emotions and project new emotions onto portrait photos.