gracezhao1997
commented
1 year ago Hi, you can freely define a feature extractor for your purpose to decide which feature you want to save. For example, when perform style transfer, you can use Gramm matrix or other pretrained feature extractor.
Thanks for your work, I would like to ask you if I need to modify the energy function if I want to perform image translation on paired images, such as the image below. In your paper two feature extractors are trained, one for the domain-independent features (pose, color, etc. on Cat → Dog) and one for the domain-specific features (beard, nose, etc. on Cat → Dog), maybe this is more about retaining the cat’s color and other features, and discarding its structural texture and other features, then it seems that my paired image translation does not need to save color, etc. features, but it is necessary to preserve features such as structure and texture, so how to achieve such an effect, I wonder if you can answer my doubts for me, thank you very much for your contribution, and hope to get your reply!
