PSGAN: Pose and Expression Robust Spatial-Aware GAN for Customizable Makeup Transfer

Abstract

Partial makeup trasnfer (lip, skin, eye) Source Reference Result Light makeup Heavy makeup Large poses and expressions differences Figure 1. Our model allows users to control both the shade of makeup and facial parts to transfer. The first row on the left shows the results of only transferring partial makeup style from the reference. The second row shows the results with controllable shades. Moreover, our method can perform makeup transfer between images that have different poses and expressions, as shown on the right part of the figure. Best viewed in color.

A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications

“…Field Subfield Method Super-resolution SRGAN [63], ESRGAN [64], Cycle-in-Cycle GANs [65], SRDGAN [66], TGAN [67] DR-GAN [68], TP-GAN [69], PG 2 [70], PSGAN [71], Image synthesis and manipulation APDrawingGAN [72], IGAN [73], Image processing and computer vision introspective adversarial networks [74], GauGAN [75] Texture synthesis MGAN [76], SGAN [77], PSGAN [78]…”

“…Siarohin et al [326] proposed deformable gans for pose-based human image generation. Pose-robust spatialaware GAN (PSGAN) for customizable makeup transfer is proposed in [71].…”

SLGAN: Style- and Latent-guided Generative Adversarial Network for Desirable Makeup Transfer and Removal

“…However, this approach often failed to transfer in-the-wild images and could not partially adjust transfers. To overcome these problems, PSGAN [16] performed a MT using the Attentive Makeup Morphing module with an attention mechanism based on spatial information, using a style-guided architecture. Furthermore, PSGAN could adjust the proportion of the style of a reference image by adjusting the weight of attention features.…”

“…These features include (1) facial components, (2) interactive color adjustments, (3) makeup variations, (4) robustness to poses and expressions, and (5) the use of multiple reference images. Several studies of makeup transfer (MT) and removal (MR) have been proposed [3,4,16,24], and most hove used GANs [2,10]. However, extant works have never striven to satisfy all five of the mentioned feature variables.…”

“…BeautyGlow [4] was limited to use of only one reference image. PS-GAN [16] was robust to changes of pose and expression, but it was not possible to perform MR. In this paper, to adequately address all five features, we propose the SLGAN.…”