As an Amazon Associate we earn from qualifying purchases.
I've been trying to wrap my head around the nuances between style transfer and style synthesis. It seems like the core difference boils down to this: style transfer takes a specific content image and a specific style image, and then merges the style of one onto the content of the other. You're essentially modifying an existing image. Style synthesis, on the other hand, seems more about creating entirely new images that embody the characteristics of a particular style, without necessarily needing a content image as input.
For example, with style transfer, you might take a photo of your dog and apply Van Gogh's "Starry Night" style to it. With style synthesis, you might train a model to generate entirely new "starry Night"-esque landscapes that never existed before.
Is that a fairly accurate high-level understanding? I'm curious to hear if anyone has more insights or practical examples that further clarify the distinction between thes two techniques. I'm also wondering how the datasets used for training differ significantly between the two. It seems a style transfer model could benefit from paired data (image and style mapping),while a synthesis model might be better trained with a large corpus demonstrating only the 'style'.