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

Abstract

Interpolation results of makeup transfer and removal. We propose a style-and latent-guided generative adversarial network, which allows the user to adjust makeup shading in an image to obtain a desirable result. Our model interpolates from light to heavy makeup based on a style-guided value with a single reference image (first row) and two reference images (second row). Our model can also arbitrarily remove makeup by modulating a latent-guided value (third row).

Makeup transfer: A review

“…The proposed Tensorflow deep learning computing framework has further advanced the field of MT [1]. For example, pedestrian re-identification [2][3][4][5][6][7], face attribute recognition [8][9][10][11], mechanical device diagnosis [12], hyperspectral image classification [13][14][15][16][17], style transfer [18][19][20][21][22][23], smart grid [24,25] etc. In addition, deep learning has also achieved good results in the field of MT, and currently, MT methods based on generative adversarial networks have become the mainstream methods in the field of makeup change.…”