891 Hits in 4.2 sec

Unsupervised Discovery of Interpretable Directions in the GAN Latent Space [article]

Andrey Voynov, Artem Babenko
2020 arXiv   pre-print
In this paper, we introduce an unsupervised method to identify interpretable directions in the latent space of a pretrained GAN model.  ...  The latent spaces of GAN models often have semantically meaningful directions.  ...  We propose the first unsupervised approach for the discovery of semantically meaningful directions in the GAN latent space.  ... 
arXiv:2002.03754v3 fatcat:iq6pttabwvavlc5qbw6ntochli

LatentCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions [article]

Oğuz Kaan Yüksel, Enis Simsar, Ezgi Gülperi Er, Pinar Yanardag
2021 arXiv   pre-print
Recent research has shown that it is possible to find interpretable directions in the latent spaces of pre-trained Generative Adversarial Networks (GANs).  ...  In this work, we propose a contrastive learning-based approach to discover semantic directions in the latent space of pre-trained GANs in a self-supervised manner.  ...  We also acknowledge the support of NVIDIA Corporation through the donation of the TITAN X GPU and GCP research credits from Google. We thank to Irem Simsar for proof-reading our paper.  ... 
arXiv:2104.00820v2 fatcat:v7gcu5mjureyfa5o2uvvfjk6pm

Closed-Form Factorization of Latent Semantics in GANs [article]

Yujun Shen, Bolei Zhou
2021 arXiv   pre-print
A rich set of interpretable dimensions has been shown to emerge in the latent space of the Generative Adversarial Networks (GANs) trained for synthesizing images.  ...  In particular, we take a closer look into the generation mechanism of GANs and further propose a closed-form factorization algorithm for latent semantic discovery by directly decomposing the pre-trained  ...  The crux of interpreting the latent space of GANs is to find the meaningful directions in the latent space corresponding to the human-understandable concepts [7, 15, 24, 22, 27] .  ... 
arXiv:2007.06600v4 fatcat:y2jtbcp345gsvad6dec7lbv6ou

Interpreting Generative Adversarial Networks for Interactive Image Generation [article]

Bolei Zhou
2022 arXiv   pre-print
This chapter gives a summary of recent works on interpreting deep generative models. The methods are categorized into the supervised, the unsupervised, and the embedding-guided approaches.  ...  Significant progress has been made by the advances in Generative Adversarial Networks (GANs) for image generation.  ...  [7] perform PCA on the sam-pled data to find primary directions in the latent space.  ... 
arXiv:2108.04896v2 fatcat:4hudnsqpyzexhi66vcbzbwxyfi

WarpedGANSpace: Finding non-linear RBF paths in GAN latent space [article]

Christos Tzelepis, Georgios Tzimiropoulos, Ioannis Patras
2021 arXiv   pre-print
This work addresses the problem of discovering, in an unsupervised manner, interpretable paths in the latent space of pretrained GANs, so as to provide an intuitive and easy way of controlling the underlying  ...  In doing so, it addresses some of the limitations of the state-of-the-art works, namely, a) that they discover directions that are independent of the latent code, i.e., paths that are linear, and b) that  ...  RPGAN: gans interpretability via random routing. CoRR, abs/1912.10920, 2019. [34] A. Voynov and A. Babenko. Unsupervised discovery of in- terpretable directions in the GAN latent space.  ... 
arXiv:2109.13357v1 fatcat:cnbdieg4rnaobfin4fmvyw2rpm

Cluster-guided Image Synthesis with Unconditional Models [article]

Markos Georgopoulos, James Oldfield, Grigorios G Chrysos, Yannis Panagakis
2021 arXiv   pre-print
Generative Adversarial Networks (GANs) are the driving force behind the state-of-the-art in image generation.  ...  In this work, we focus on controllable image generation by leveraging GANs that are well-trained in an unsupervised fashion.  ...  Unsupervised discovery of interpretable directions in the gan latent space. arXiv preprint arXiv:2002.03754, 2020. 2 [40] Xueting Yan, Ishan Misra, Abhinav Gupta, Deepti Ghadi- yaram, and  ... 
arXiv:2112.12911v1 fatcat:l3zhx7bxsjaohertladvyqbbna

Disentangled Representations from Non-Disentangled Models [article]

Valentin Khrulkov, Leyla Mirvakhabova, Ivan Oseledets, Artem Babenko
2021 arXiv   pre-print
The dominating paradigm of unsupervised disentanglement is currently to train a generative model that separates different factors of variation in its latent space.  ...  Constructing disentangled representations is known to be a difficult task, especially in the unsupervised scenario.  ...  First, we search for a set of k orthogonal interpretable directions in the latent space of the pretrained GAN in an unsupervised manner.  ... 
arXiv:2102.06204v1 fatcat:yeg24kjkkzdnhhcxv4igm5mz3u

Unsupervised Representation Adversarial Learning Network: from Reconstruction to Generation [article]

Yuqian Zhou, Kuangxiao Gu, Thomas Huang
2019 arXiv   pre-print
In theory, we minimize the upper bound of the two conditional entropy loss between the latent variables and the observations together to achieve the cycle consistency.  ...  This paper aims at learning a disentangled representation effective for all of them in an unsupervised way.  ...  But it does not learn a disentangled latent space for semantic interpretation and knowledge discovery.  ... 
arXiv:1804.07353v2 fatcat:t3qolvtdbbhxhmk45cjwrbdxky

GAN "Steerability" without optimization [article]

Nurit Spingarn-Eliezer, Ron Banner, Tomer Michaeli
2021 arXiv   pre-print
Recent research has shown remarkable success in revealing "steering" directions in the latent spaces of pre-trained GANs.  ...  This applies to user-prescribed geometric transformations, as well as to unsupervised discovery of more complex effects.  ...  Unsupervised discovery of interpretable directions in the gan latent space. arXiv preprint arXiv:2002.03754, 2020. Tom White. Sampling generative networks. arXiv preprint arXiv:1609.04468, 2016.  ... 
arXiv:2012.05328v2 fatcat:azj2jj7zj5fhfpeqtjzf3juzd4

Enjoy Your Editing: Controllable GANs for Image Editing via Latent Space Navigation [article]

Peiye Zhuang, Oluwasanmi Koyejo, Alexander G. Schwing
2021 arXiv   pre-print
Classic approaches for this task use a Generative Adversarial Net (GAN) to learn a latent space and suitable latent-space transformations.  ...  loss that encourages the maintenance of image identity and photo-realism.  ...  Note that work by Voynov & Babenko (2020) is unsupervised, i.e., the latent-space directions are human interpreted.  ... 
arXiv:2102.01187v3 fatcat:jsprls7hcjcv7bu6i33vszpbia

Fantastic Style Channels and Where to Find Them: A Submodular Framework for Discovering Diverse Directions in GANs [article]

Enis Simsar and Umut Kocasari and Ezgi Gülperi Er and Pinar Yanardag
2022 arXiv   pre-print
The discovery of interpretable directions in the latent spaces of pre-trained GAN models has recently become a popular topic.  ...  In this study, we design a novel submodular framework that finds the most representative and diverse subset of directions in the latent space of StyleGAN2.  ...  directions. [36] uses a self-supervised contrastive learning based method to discover interpretable directions in the latent space of pre-trained BigGAN and StyleGAN2 models.  ... 
arXiv:2203.08516v2 fatcat:svome3s37vh3rluiv5yroxthrm

The Geometry of Deep Generative Image Models and its Applications [article]

Binxu Wang, Carlos R. Ponce
2021 arXiv   pre-print
GAN inversion) and facilitates unsupervised discovery of interpretable axes.  ...  This geometric understanding unifies key previous results related to GAN interpretability. We show that the use of this metric allows for more efficient optimization in the latent space (e.g.  ...  We thank Hao Sun (CUHK) in providing experience for the submission and rebuttal process.  ... 
arXiv:2101.06006v2 fatcat:ngg63aohlbc47elmt353f6elbu

Unsupervised Primitive Discovery for Improved 3D Generative Modeling

Salman H. Khan, Yulan Guo, Munawar Hayat, Nick Barnes
2019 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)  
3D shape generation is a challenging problem due to the high-dimensional output space and complex part configurations of real-world objects.  ...  To this end, we introduce an unsupervised primitive discovery algorithm based on a higher-order conditional random field model.  ...  , our model is jointly trained on all shape classes, (b) It provides better interpretability of generator's latent space and can incorporate user input to generate desired shapes, (c) The learned model  ... 
doi:10.1109/cvpr.2019.00997 dblp:conf/cvpr/KhanGHB19 fatcat:bepx6sunpjduvhrgha5ubpvzki

Quantitative comparison of principal component analysis and unsupervised deep learning using variational autoencoders for shape analysis of motile cells [article]

Caleb K. Chan, Amalia Hadjitheodorou, Tony Y.-C. Tsai, Julie A. Theriot
2020 bioRxiv   pre-print
Furthermore, by including cell speed into the training of the VAE-GAN, we were able to incorporate cell shape and speed into the same latent space.  ...  Contrary to the conventional viewpoint that the latent space is a "black box", we demonstrated that the information learned and encoded within the latent space is consistent with PCA and is reproducible  ...  latent space have no direct physical meaning and are not arranged in any meaningful 408 order, interpreting the biological significance of positional variation within the latent 409 space is much less  ... 
doi:10.1101/2020.06.26.174474 fatcat:ovabyy3vj5eoteyrr7m3wkhdza

Learning Disentangled Representation by Exploiting Pretrained Generative Models: A Contrastive Learning View [article]

Xuanchi Ren, Tao Yang, Yuwang Wang, Wenjun Zeng
2022 arXiv   pre-print
For the generative models pretrained without any disentanglement term, the generated images show semantically meaningful variations when traversing along different directions in the latent space.  ...  DisCo achieves the state-of-the-art disentangled representation learning and distinct direction discovering, given pretrained non-disentangled generative models including GAN, VAE, and Flow.  ...  method to explore interpretable directions in the latent space of a pretrained GAN.  ... 
arXiv:2102.10543v2 fatcat:o5733u4egfhsjjocoj2gerayw4
« Previous Showing results 1 — 15 out of 891 results