TransEditor: Transformer-Based Dual-Space GAN for Highly Controllable Facial Editing

Yang, Xu; Yin, Yueqin; Jiang, Liming; Wu, Qianyi; Chen, Zheng; Loy, Chen Change; Dai, Bo; Wu, Wayne

doi:10.1109/cvpr52688.2022.00753

Cited by 39 publications

(8 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Current CycleGAN [17] implements complex local texture transformations [28] on unpaired image generation tasks in two domains, such as semantic synthesis of images [29,30] (horse to zebra transfor), style transfer [31,32] (season exchange), and image coloring [33,34] (coloring of centuries-old photos). This section adopts the generator and discriminator of CycleGAN [17] as the basic generative adversarial network.…”

Section: Generative Adversarial Network For Multi-domain Feature Fusionmentioning

confidence: 99%

Content and style transfer with generative adversarial network

Ding

Hou

et al. 2023

Preprint

View full text Add to dashboard Cite

The generative adversarial network is often used for image conversion tasks such as image coloring, semantic composition,style transfer, etc.,but at this stage, the training of image generation models often depends on a large number of paired datasets, and can only achieve conversion between two image domains.When processing tasks in more than two domains,it lacks scalability and robustness.To solve the above problems,this paper proposes a Content and Style transfer model based on Generative Adversarial Network (CS-GAN).This model can fuse style features (such as monet style,cubism) and content features (such as color ,texture) of fashion items on unpaired datasets at the same time,which can realize the conversion of multiple image domains,so as to effectively complete the task of transferring the content and style of fashion items.In particular,we propose a layer consistent dynamic convolution (LCDC) method,which encodes the style image as a learnable convolution parameter,which can adaptively learn style features,and more flexibly and efficiently complete the arbitrary style transfer of fashion items.To validate the performance of our model,we conducts comparative experiments and results analysis on the public fashion dataset. Compared with other mainstream methods,this method has improved in image synthesis quality,Inception Score (IS) and Frechet Inception Dinstance score (FID) evaluation index.

show abstract

Section: Generative Adversarial Network For Multi-domain Feature Fusionmentioning

confidence: 99%

Content and style transfer with generative adversarial network

Ding

Hou

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…Barbershop [37] combined the latent code and mask to edit the specified region. Transeditor [34] controlled the face pose and face style individually by mapping the face into the double space and found the average vector between the latent code of the two faces for linear transformation to achieve facial attribute editing. The optimization-based approach is similar to the target-specific face-swapping algorithm, which is more advantageous in generation quality but less efficient.…”

Section: Latent Space Manipulationmentioning

confidence: 99%

End-to-end Face-swapping via Adaptive Latent Representation Learning

Lin¹,

Hu²,

Shen³

et al. 2023

Preprint

View full text Add to dashboard Cite

“…In [151], DAT further makes z c to have a symmetric structure similar to z and controls the feature map at each scale. Recently, TransEditor [152] incorporates transformer blocks to establish the interaction between two latent spaces, which improves the controllability and flexibility of FAM.…”

Section: Binary Semantic Decompositionmentioning

confidence: 99%

GAN-Based Face Attribute Editing

Liu

Cao

et al. 2020

IEEE Access

View full text Add to dashboard Cite

Recently, a variety of methods using the Generative Adversarial Network (GAN) for face editing have been proposed. However, the existing methods cannot control the editing content of the face elements according to the user-specified attributes or need to train a conditional GAN for editing tasks, which means it is difficult to add new attributes in the future. In this paper, a method to edit face attributes by editing the latent variable with the help of a pre-trained unconditional GAN and a linear classification model is proposed. In particular, face attribute editing is divided into two separate stages: Firstly, based on the optimization function, the generative model does a latent variable search to generate a high-quality face image that is similar to the input image. Secondly, by editing the latent variable of the GAN, the attribute of the generated face image can be modified indirectly, so it is almost unaffected by the training process and network structure of GAN, which means it is a flexible method for nearly all GAN network. Images of the FFHQ dataset are edited by attribute labels defined in Celeba dataset for experiments. These experiments prove that our method can edit a variety of face images that vary with race, gender, age, and camera shooting angle. The overall quality of the edited image is not inferior to the other face attribute editing method, and attribute classification for edited image shows a 92.6% attribute editing success rate of the proposed method.

show abstract

TransEditor: Transformer-Based Dual-Space GAN for Highly Controllable Facial Editing

Cited by 39 publications

References 29 publications

Content and style transfer with generative adversarial network

Content and style transfer with generative adversarial network

End-to-end Face-swapping via Adaptive Latent Representation Learning

GAN-Based Face Attribute Editing

Contact Info

Product

Resources

About