Few-shot Semantic Image Synthesis Using StyleGAN Prior
This work solves the problem of costly pixel-wise annotation for image synthesis, offering a practical solution for applications like content creation, though it is incremental by building on existing StyleGAN methods.
The paper addresses generating photorealistic images from semantic layouts with limited annotated data by using a StyleGAN prior to create pseudo semantic masks, enabling high-quality synthesis from dense or sparse inputs and showing improved layout fidelity and visual quality in few-shot settings.
This paper tackles a challenging problem of generating photorealistic images from semantic layouts in few-shot scenarios where annotated training pairs are hardly available but pixel-wise annotation is quite costly. We present a training strategy that performs pseudo labeling of semantic masks using the StyleGAN prior. Our key idea is to construct a simple mapping between the StyleGAN feature and each semantic class from a few examples of semantic masks. With such mappings, we can generate an unlimited number of pseudo semantic masks from random noise to train an encoder for controlling a pre-trained StyleGAN generator. Although the pseudo semantic masks might be too coarse for previous approaches that require pixel-aligned masks, our framework can synthesize high-quality images from not only dense semantic masks but also sparse inputs such as landmarks and scribbles. Qualitative and quantitative results with various datasets demonstrate improvement over previous approaches with respect to layout fidelity and visual quality in as few as one- or five-shot settings.