Do you condition the generative model (let's say, VAE) on the task identity or the class label or both? If I condition the VAE on both task identity and class label, then I have to provide both the two information to the VAE when testing, right? Why I'm asking this is that the biggest difference for task-incremental and class-incremental in continual image classification is whether the model is aware of the task identity during test phase, then what is the detailed experimental setup for class-incremental continual image generation?
Asked
Active
Viewed 5 times
