Fashion++: Minimal Edits for Outfit Improvement

Abstract

Given an outfit, what small changes would most improve its fashionability? This question presents an intriguing new vision challenge. We introduce Fashion++, an approach that proposes minimal adjustments to a full-body clothing outfit that will have maximal impact on its fashionability. Our model consists of a deep image generation neural network that learns to synthesize clothing conditioned on learned per-garment encodings. The latent encodings are explicitly factorized according to shape and texture, thereby allowing direct edits for both fit/presentation and color/patterns/material, respectively. We show how to bootstrap Web photos to automatically train a fashionability model, and develop an activation maximization-style approach to transform the input image into its more fashionable self. The edits suggested range from swapping in a new garment to tweaking its color, how it is worn (e.g., rolling up sleeves), or its fit (e.g., making pants baggier). Experiments demonstrate that Fashion++ provides successful edits, both according to automated metrics and human opinion. Project

Fashion Meets Computer Vision

“…They presented an end-to-end multi-layered comparison network to predict the compatibility between different items at different layers and use the backpropagation gradient for diagnosis. Hsiao et al [65] proposed Fashion++ to make minimal adjustments to a full-body clothing outfit that have a maximal impact on its fashionability. Besides, Song et al [169] took user preferences into account to present a personalized compatibility modeling scheme GP-BPR.…”

Mt-Gan: A Training Framework To Enhance Image Classification Task With Image Translation

“…GANs have demonstrated their success in various image synthesis applications [11] such as fashion image synthesis [12] and conditional image synthesis [13]. Recent advances of using GAN for image generation or translation have shown great progress in terms of improved image quality with both higher resolution [14] and higher fidelity [15] [16] [17].…”

From Design Draft to Real Attire: Unaligned Fashion Image Translation

“…FashionGAN [30] and AMGAN [1] are introduced to conduct text-guided fashion item manipulation such as long sleeves to short sleeves while preserving other parts of the clothes. Fashion++ [6], VITON [5], CP-VITON [22] and E2E [19,20] can make adjustments to clothing outfit and realize the function of trying on clothes with one photo.…”

“…Thanks to the recent development of generative adversarial networks (GANs), the intelligent image-based fashion editing becomes possible. The model can automatically generate the corresponding clothing according to the user's language guidance [1,30], or recommend the clothing matching according to the user's favorite [10], or even provide virtual try-on service [5,6,22], etc. These models enable normal users to convert their creative ideas into corresponding fashion items, which greatly reduces the difficulty for user customization, and is of certain entertainment value and huge commercial value.…”