Controlling Generative Networks

Erik explains the historical approaches to controlling generative adversarial networks, including labeling training data and using pretrained generators. However, their work takes a different approach, asking the model to show them the most important directions of change, resulting in a plethora of interesting sliders to explore.