Closed OwalnutO closed 6 years ago
We didn't really perform any experiments with increased image resolution. But my intuition says simply increasing the image size would result in performance deterioration. I think you'd have to correspondingly increase the dataset size. Also, I think it'll require some hyperparameter tuning. But I think you should observe better performance once you make these changes. In order to get conditional generation in a DeliGAN setting, just take any conditional GAN framework, and sample from the mixture of Gaussian setup as in DeliGAN, instead of using a unit normal distribution . This was the point of DeliGAN. View it more like a layer that can be plugged into any other architecture to learn a more flexible prior distribution for the generator. You could also tweak the DeliGAN model we have provided to emulate the conditional GAN setup, but that would be more cumbersome.
Two questions: 1.Do you find performance reduction if you increase the size of the generated image? Such as the dataset of the sketch? Can you give some suggestions if I want to increase the size of the results? 2.How to modify the network if I want to generate specific kind of images, just like the conditional GAN? Any suggestions? Thanks!