191 Hits in 3.5 sec

Unsupervised Disentanglement of Linear-Encoded Facial Semantics [article]

Yutong Zheng, Yu-Kai Huang, Ran Tao, Zhiqiang Shen, Marios Savvides
2021 arXiv   pre-print
We propose a method to disentangle linear-encoded facial semantics from StyleGAN without external supervision.  ...  The method derives from linear regression and sparse representation learning concepts to make the disentangled latent representations easily interpreted as well.  ...  Conclusion We have presented an unsupervised learning framework for disentangling linear-encoded facial semantics from StyleGAN.  ... 
arXiv:2103.16605v1 fatcat:yhddi74evzan3h4hd7xjnd2caa

State-of-the-Art in the Architecture, Methods and Applications of StyleGAN [article]

Amit H. Bermano and Rinon Gal and Yuval Alaluf and Ron Mokady and Yotam Nitzan and Omer Tov and Or Patashnik and Daniel Cohen-Or
2022 arXiv   pre-print
Among StyleGAN's most interesting aspects is its learned latent space. Despite being learned with no supervision, it is surprisingly well-behaved and remarkably disentangled.  ...  However, the control offered by StyleGAN is inherently limited to the generator's learned distribution, and can only be applied to images generated by StyleGAN itself.  ...  direction resides in extracting knowledge from StyleGAN for non-generative needs.  ... 
arXiv:2202.14020v1 fatcat:qu3plbdnszdujcwxwq3zizysje

A Latent Transformer for Disentangled Face Editing in Images and Videos [article]

Xu Yao, Alasdair Newson, Yann Gousseau, Pierre Hellier
2021 arXiv   pre-print
Our model achieves a disentangled, controllable, and identity-preserving facial attribute editing, even in the challenging case of real (i.e., non-synthetic) images and videos.  ...  To tackle these limitations, we propose to edit facial attributes via the latent space of a StyleGAN generator, by training a dedicated latent transformation network and incorporating explicit disentanglement  ...  [44] proposed to navigate the latent space of StyleGAN in a non-linear manner to achieve disentangled manipulations of facial attributes.  ... 
arXiv:2106.11895v2 fatcat:czi7onsp75e3bm4zontcks7or4

Only a Matter of Style: Age Transformation Using a Style-Based Regression Model [article]

Yuval Alaluf, Or Patashnik, Daniel Cohen-Or
2021 arXiv   pre-print
Moreover, unlike approaches that operate solely in the latent space using a prior on the path controlling age, our method learns a more disentangled, non-linear path.  ...  Finally, we demonstrate that the end-to-end nature of our approach, coupled with the rich semantic latent space of StyleGAN, allows for further editing of the generated images.  ...  We would also like to thank the anonymous reviewers for their insightful comments and constructive remarks.  ... 
arXiv:2102.02754v2 fatcat:3ut6yy66kjcyhcbc7rauw2cgxq

CLIP2StyleGAN: Unsupervised Extraction of StyleGAN Edit Directions [article]

Rameen Abdal, Peihao Zhu, John Femiani, Niloy J. Mitra, Peter Wonka
2021 arXiv   pre-print
We evaluate the effectiveness of the proposed method and demonstrate that extraction of disentangled labeled StyleGAN edit directions is indeed possible, and reveals interesting and non-trivial edit directions  ...  The success of StyleGAN has enabled unprecedented semantic editing capabilities, on both synthesized and real images.  ...  Image editing frameworks in the StyleGAN domain [3, 21, 39, 43] analyze the latent space to identify linear and non-linear paths for semantic editing.  ... 
arXiv:2112.05219v1 fatcat:o5ozhktixrgnpphjgy5gix5gmy

InterFaceGAN: Interpreting the Disentangled Face Representation Learned by GANs [article]

Yujun Shen, Ceyuan Yang, Xiaoou Tang, Bolei Zhou
2020 arXiv   pre-print
We first find that GANs learn various semantics in some linear subspaces of the latent space.  ...  Extensive experimental results suggest that learning to synthesize faces spontaneously brings a disentangled and controllable face representation.  ...  model weights for gradual editing.  ... 
arXiv:2005.09635v2 fatcat:ahflrw2fkjafnp4fjjymo7cjae

Interpreting the Latent Space of GANs for Semantic Face Editing [article]

Yujun Shen, Jinjin Gu, Xiaoou Tang, Bolei Zhou
2020 arXiv   pre-print
In this work, we propose a novel framework, called InterFaceGAN, for semantic face editing by interpreting the latent semantics learned by GANs.  ...  We find that the latent code of well-trained generative models actually learns a disentangled representation after linear transformations.  ...  After learning such a non-linear mapping, GAN is capable of producing photo-realistic images from randomly sampled latent codes.  ... 
arXiv:1907.10786v3 fatcat:mzfdqkijtnhurfuirh7v2us634

Editing in Style: Uncovering the Local Semantics of GANs [article]

Edo Collins, Raja Bala, Bob Price, Sabine Süsstrunk
2020 arXiv   pre-print
Instead, it relies on the emergent disentanglement of semantic objects that is learned by StyleGAN during its training.  ...  Focusing on StyleGAN, we introduce a simple and effective method for making local, semantically-aware edits to a target output image.  ...  In this paper we show that deep generative models like PG-GAN, StyleGAN and the recent StyleGAN2 [17] learn a representation of objects and object-parts that is disentangled in the sense that various  ... 
arXiv:2004.14367v2 fatcat:pwlkop4hezcgtk4jj6gju63wsq

SemanticStyleGAN: Learning Compositional Generative Priors for Controllable Image Synthesis and Editing [article]

Yichun Shi, Xiao Yang, Yangyue Wan, Xiaohui Shen
2022 arXiv   pre-print
When combined with editing methods designed for StyleGANs, it can achieve a more fine-grained control to edit synthesized or real images.  ...  Recent studies have shown that StyleGANs provide promising prior models for downstream tasks on image synthesis and editing.  ...  In StyleGANs [36, 37] , to handle the non-linearity of data distribution, z is first mapped into a latent code w ∼ W with an MLP.  ... 
arXiv:2112.02236v3 fatcat:dvwd332znreenmgb6b4mipf6ia

Multi-Attribute Balanced Sampling for Disentangled GAN Controls [article]

Perla Doubinsky
2022 arXiv   pre-print
We demonstrate the effectiveness of this approach by extracting disentangled linear directions for face manipulation on two popular GAN architectures, PGGAN and StyleGAN, and two datasets, CelebAHQ and  ...  We show that this approach outperforms state-of-the-art classifier-based methods while avoiding the need for disentanglement-enforcing post-processing.  ...  According to Table 1 (c), our approach succeeds to extract directions allowing disentangled edits without requiring conditional manipulation.  ... 
arXiv:2111.00909v2 fatcat:gti3cqoixjgd7kqehp6na5soty

StyleFlow: Attribute-conditioned Exploration of StyleGAN-Generated Images using Conditional Continuous Normalizing Flows [article]

Rameen Abdal, Peihao Zhu, Niloy Mitra, Peter Wonka
2020 arXiv   pre-print
We evaluate our method using the face and the car latent space of StyleGAN, and demonstrate fine-grained disentangled edits along various attributes on both real photographs and StyleGAN generated images  ...  For example, for faces, we vary camera pose, illumination variation, expression, facial hair, gender, and age.  ...  Firstly, a ribute-guided edits amount to non-linear curves in the StyleGAN latent space.  ... 
arXiv:2008.02401v2 fatcat:wbzz4t23uvcozk3nav675heewm

BSD-GAN: Branched Generative Adversarial Network for Scale-Disentangled Representation Learning and Image Synthesis [article]

Zili Yi, Zhiqin Chen, Hao Cai, Wendong Mao, Minglun Gong, Hao Zhang
2020 arXiv   pre-print
a wide range of generation and editing tasks.  ...  We introduce BSD-GAN, a novel multi-branch and scale-disentangled training method which enables unconditional Generative Adversarial Networks (GANs) to learn image representations at multiple scales, benefiting  ...  (0.0, 10.5) for StyleGAN respectively).  ... 
arXiv:1803.08467v5 fatcat:ft4atylkxbdbhey6uwojjby53q

FEAT: Face Editing with Attention [article]

Xianxu Hou, Linlin Shen, Or Patashnik, Daniel Cohen-Or, Hui Huang
2022 arXiv   pre-print
The guidance for the latent space edits is achieved by employing CLIP, which has recently been shown to be effective for text-driven edits.  ...  In this paper, we build on the StyleGAN generator, and present a method that explicitly encourages face manipulation to focus on the intended regions by incorporating learned attention maps.  ...  InterfaceGAN performs linear manipulation in the GAN latent space, while StyleFlow extracts non-linear editing paths in the latent space using conditional continuous normalizing flows.  ... 
arXiv:2202.02713v1 fatcat:sahccqhfqzcs3fqeligcepbf64

Hijack-GAN: Unintended-Use of Pretrained, Black-Box GANs [article]

Hui-Po Wang, Ning Yu, Mario Fritz
2021 arXiv   pre-print
We achieve this by an iterative scheme that also allows gaining control over the image generation process despite the highly non-linear latent spaces of the latest GAN models.  ...  We show that state-of-the-art GAN models -- such as they are being publicly released by researchers and industry -- can be used for a range of applications beyond unconditional image generation.  ...  non-linear latent spaces (e.g., Z-space of StyleGAN [25, 33] ).  ... 
arXiv:2011.14107v2 fatcat:6gz6vsw7kjhypj6m5oswmsop5e

Towards Disentangling Latent Space for Unsupervised Semantic Face Editing [article]

Kanglin Liu and Gaofeng Cao and Fei Zhou and Bozhi Liu and Jiang Duan and Guoping Qiu
2021 arXiv   pre-print
Therefore, unsupervised attribute editing in an disentangled latent space is key to performing neat and versatile semantic face editing.  ...  In this paper, we present a new technique termed Structure-Texture Independent Architecture with Weight Decomposition and Orthogonal Regularization (STIA-WO) to disentangle the latent space for unsupervised  ...  Even though (3) is a linear equation, there is a non-linear relationship betweenŴ and S.  ... 
arXiv:2011.02638v2 fatcat:ygf5pkylbvdrxcxytekgxv452i
« Previous Showing results 1 — 15 out of 191 results