Pixel-Level Domain Transfer

TL;DR

This paper introduces a pixel-level domain transfer model that generates realistic images in a target domain from an input image, using adversarial training and a novel domain discriminator, demonstrated on clothing images.

Contribution

The paper proposes a new image generation model that transfers images between domains at the pixel level with a domain relevance discriminator, advancing domain transfer techniques.

Findings

Successfully generated clothing images from input images of dressed persons.

Created a high-quality clothing dataset for domain transfer tasks.

Achieved realistic and relevant image generation results.

Abstract

We present an image-conditional image generation model. The model transfers an input domain to a target domain in semantic level, and generates the target image in pixel level. To generate realistic target images, we employ the real/fake-discriminator as in Generative Adversarial Nets, but also introduce a novel domain-discriminator to make the generated image relevant to the input image. We verify our model through a challenging task of generating a piece of clothing from an input image of a dressed person. We present a high quality clothing dataset containing the two domains, and succeed in demonstrating decent results.