Timezone: »

A Loss Function for Generative Neural Networks Based on Watson’s Perceptual Model
Steffen Czolbe · Oswin Krause · Ingemar Cox · Christian Igel

Tue Dec 08 09:00 AM -- 11:00 AM (PST) @ Poster Session 1 #409

To train Variational Autoencoders (VAEs) to generate realistic imagery requires a loss function that reflects human perception of image similarity. We propose such a loss function based on Watson's perceptual model, which computes a weighted distance in frequency space and accounts for luminance and contrast masking. We extend the model to color images, increase its robustness to translation by using the Fourier Transform, remove artifacts due to splitting the image into blocks, and make it differentiable. In experiments, VAEs trained with the new loss function generated realistic, high-quality image samples. Compared to using the Euclidean distance and the Structural Similarity Index, the images were less blurry; compared to deep neural network based losses, the new approach required less computational resources and generated images with less artifacts.

Author Information

Steffen Czolbe (University of Copenhagen)
Oswin Krause (University of Copenhagen)
Ingemar Cox (University College London)
Christian Igel (University of Copenhagen)

More from the Same Authors