StyleGAN2 Distillation for Feed-forward Image Manipulation (2020-03-07T00:00:00.000000Z)

TL;DR

It is shown that the quality of generation using the proposed pipeline is comparable to StyleGAN2 backpropagation and current state-of-the-art methods in these particular tasks, and the resulting pipeline is an alternative to existing GANs, trained on unpaired data.

Abstract

StyleGAN2 is a state-of-the-art network in generating realistic images. Besides, it was explicitly trained to have disentangled directions in latent space, which allows efficient image manipulation by varying latent factors. Editing existing images requires embedding a given image into the latent space of StyleGAN2. Latent code optimization via backpropagation is commonly used for qualitative embedding of real world images, although it is prohibitively slow for many applications. We propose a way to distill a particular image manipulation of StyleGAN2 into image-to-image network trained in paired way. The resulting pipeline is an alternative to existing GANs, trained on unpaired data. We provide results of human faces' transformation: gender swap, aging/rejuvenation, style transfer and image morphing. We show that the quality of generation using our method is comparable to StyleGAN2 backpropagation and current state-of-the-art methods in these particular tasks.

Authors

Yuri Viazovetskyi

1 papers

V. Ivashkin

1 papers

Evgenii Kashin

1 papers

TL;DR

Abstract

Authors

References63 items

StarGAN v2: Diverse Image Synthesis for Multiple Domains

Analyzing and Improving the Image Quality of StyleGAN

Image2StyleGAN++: How to Edit the Embedded Images?

Self-Training With Noisy Student Improves ImageNet Classification

Few-shot Video-to-Video Synthesis

S2GAN: Share Aging Factors Across Ages and Share Aging Trends Among Individuals

Face-to-Parameter Translation for Game Character Auto-Creation

Interpreting the Latent Space of GANs for Semantic Face Editing

On the "steerability" of generative adversarial networks

GANalyze: Toward Visual Definitions of Cognitive Image Properties

Style Generator Inversion for Image Enhancement and Animation

Classification Accuracy Score for Conditional Generative Models

Few-Shot Unsupervised Image-to-Image Translation

DRIT++: Diverse Image-to-Image Translation via Disentangled Representations

Meta-Sim: Learning to Generate Synthetic Datasets

Image2StyleGAN: How to Embed Images Into the StyleGAN Latent Space?

Seeing is Not Necessarily Believing: Limitations of BigGANs for Data Augmentation

Semantic Image Synthesis With Spatially-Adaptive Normalization

Improved Knowledge Distillation via Teacher Assistant: Bridging the Gap Between Student and Teacher

Compressing GANs using Knowledge Distillation

Learning Continuous Face Age Progression: A Pyramid of GANs

A Style-Based Generator Architecture for Generative Adversarial Networks

Learning To Simulate

Large Scale GAN Training for High Fidelity Natural Image Synthesis

Attribute-Aware Face Aging With Wavelet-Based Generative Adversarial Networks

Everybody Dance Now

Video-to-Video Synthesis

Dual Conditional GANs for Face Aging and Rejuvenation

Face Aging with Identity-Preserved Conditional Generative Adversarial Networks

Multimodal Unsupervised Image-to-Image Translation

Learning to Adapt Structured Output Space for Semantic Segmentation

Global and Local Consistent Age Generative Adversarial Networks

High-Resolution Image Synthesis and Semantic Manipulation with Conditional GANs

StarGAN: Unified Generative Adversarial Networks for Multi-domain Image-to-Image Translation

CyCADA: Cycle-Consistent Adversarial Domain Adaptation

Toward Multimodal Image-to-Image Translation

Progressive Growing of GANs for Improved Quality, Stability, and Variation

Makeup-Go: Blind Reversion of Portrait Edit

Photographic Image Synthesis with Cascaded Refinement Networks

AgeDB: The First Manually Collected, In-the-Wild Age Database

GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium

Self-ensembling for visual domain adaptation

DualGAN: Unsupervised Dual Learning for Image-to-Image Translation

Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks

Arbitrary Style Transfer in Real-Time with Adaptive Instance Normalization

Domain randomization for transferring deep neural networks from simulation to the real world

Unsupervised Image-to-Image Translation Networks

Age Progression/Regression by Conditional Adversarial Autoencoder

Learning from Simulated and Unsupervised Images through Adversarial Training

Image-to-Image Translation with Conditional Adversarial Networks

Understanding and Predicting Image Memorability at a Large Scale

Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks

Distilling the Knowledge in a Neural Network

Cross-Age Reference Coding for Age-Invariant Face Recognition and Retrieval

Generative Adversarial Nets

Do Deep Nets Really Need to be Deep?

ImageNet: A large-scale hierarchical image database

Maximum Likelihood Estimation of Observer Error‐Rates Using the EM Algorithm

Stylegan encoder - converts real images to latent space

Stylegan encoder for official tensorflow implementation

Find the center of every class C k = 1 n c = k (cid:80) c ( w i )= k w i and the transition vectors from one class to another ∆ c i ,c j = C j − C i

Filter out images where faces were detected with low conﬁdence 5 . Then select only images with high classiﬁcation certainty

StyleGAN2 Distillation for Feed-forward Image Manipulation 17

Field of Study

Venue Information

Name

Type

URL

Alternate Names