Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

COutfitGAN: Learning to Synthesize Compatible Outfits Supervised by Silhouette Masks and Fashion Styles

Dongliang Zhou, Haijun Zhang, Qun Li, Jianghong Ma, Xiaofei Xu

TL;DR

This work addresses the problem of generating complete, compatible outfits conditioned on an arbitrary set of input fashion items. It proposes COutfitGAN, which combines a pyramid style extractor (mapping inputs into multi-scale $W^+$-space style embeddings) with a StyleGAN2-inspired progressive outfit generator, a UNet-based discriminator for realism, and a collocation discriminator enforcing compatibility via contrastive learning. The approach is supported by OutfitSet, a large-scale dataset of 200K outfits (800K items) with silhouette masks, enabling evaluation across varying numbers of given items. Experiments demonstrate state-of-the-art performance in similarity, authenticity, and compatibility, underscoring the method’s potential for design-oriented fashion generation and interactive outfit synthesis.

Abstract

How to recommend outfits has gained considerable attention in both academia and industry in recent years. Many studies have been carried out regarding fashion compatibility learning, to determine whether the fashion items in an outfit are compatible or not. These methods mainly focus on evaluating the compatibility of existing outfits and rarely consider applying such knowledge to 'design' new fashion items. We propose the new task of generating complementary and compatible fashion items based on an arbitrary number of given fashion items. In particular, given some fashion items that can make up an outfit, the aim of this paper is to synthesize photo-realistic images of other, complementary, fashion items that are compatible with the given ones. To achieve this, we propose an outfit generation framework, referred to as COutfitGAN, which includes a pyramid style extractor, an outfit generator, a UNet-based real/fake discriminator, and a collocation discriminator. To train and evaluate this framework, we collected a large-scale fashion outfit dataset with over 200K outfits and 800K fashion items from the Internet. Extensive experiments show that COutfitGAN outperforms other baselines in terms of similarity, authenticity, and compatibility measurements.

COutfitGAN: Learning to Synthesize Compatible Outfits Supervised by Silhouette Masks and Fashion Styles

TL;DR

This work addresses the problem of generating complete, compatible outfits conditioned on an arbitrary set of input fashion items. It proposes COutfitGAN, which combines a pyramid style extractor (mapping inputs into multi-scale -space style embeddings) with a StyleGAN2-inspired progressive outfit generator, a UNet-based discriminator for realism, and a collocation discriminator enforcing compatibility via contrastive learning. The approach is supported by OutfitSet, a large-scale dataset of 200K outfits (800K items) with silhouette masks, enabling evaluation across varying numbers of given items. Experiments demonstrate state-of-the-art performance in similarity, authenticity, and compatibility, underscoring the method’s potential for design-oriented fashion generation and interactive outfit synthesis.

Abstract

How to recommend outfits has gained considerable attention in both academia and industry in recent years. Many studies have been carried out regarding fashion compatibility learning, to determine whether the fashion items in an outfit are compatible or not. These methods mainly focus on evaluating the compatibility of existing outfits and rarely consider applying such knowledge to 'design' new fashion items. We propose the new task of generating complementary and compatible fashion items based on an arbitrary number of given fashion items. In particular, given some fashion items that can make up an outfit, the aim of this paper is to synthesize photo-realistic images of other, complementary, fashion items that are compatible with the given ones. To achieve this, we propose an outfit generation framework, referred to as COutfitGAN, which includes a pyramid style extractor, an outfit generator, a UNet-based real/fake discriminator, and a collocation discriminator. To train and evaluate this framework, we collected a large-scale fashion outfit dataset with over 200K outfits and 800K fashion items from the Internet. Extensive experiments show that COutfitGAN outperforms other baselines in terms of similarity, authenticity, and compatibility measurements.

Paper Structure

This paper contains 28 sections, 19 equations, 10 figures, 8 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Problem Formulation
  5. Motivation of Our Method
  6. An Overview of COutfitGAN
  7. The Pyramid Style Extractor
  8. The Outfit Generator
  9. The Real/Fake Discriminator
  10. The Collocation Discriminator
  11. Training losses
  12. The Adversarial Training Process
  13. Experiments
  14. The Dataset
  15. Experimental Setup and Parameter Settings
  16. ...and 13 more sections

Figures (10)

  • Figure 1: Generating outfits compatible with arbitrary fashion items.
  • Figure 2: Overview of our COutfitGAN framework.
  • Figure 3: The architecture of the pyramid style extractor.
  • Figure 4: The architecture of the outfit generator: (a) structure overview, and (b) silhouette and style (S--S) fusion block.
  • Figure 5: The UNet-based real/fake discriminator.
  • ...and 5 more figures