Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks
Непарный перенос изображений с помощью циклически согласованных состязательных сетей
2017-10-01
SCID: 54.1/pq4me6ka
Discuss with AI
adversarial losscycle consistency losscycle-consistent adversarial networksdomain mapping G and inverse Funpaired image-to-image translation
Figures from the paper
Abstract (AI)
Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain X to a target domain Y in the absence of paired examples. Our goal is to learn a mapping G : X → Y such that the distribution of images from G(X) is indistinguishable from the distribution Y using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping F : Y → X and introduce a cycle consistency loss to push F(G(X)) ≈ X (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach.
Key Findings
1
Combined the forward mapping G and inverse mapping F: Y→X with a cycle consistency loss enforcing F(G(X)) ≈ X and G(F(Y)) ≈ Y to reduce under-constrained solutions.
2
Demonstrated qualitative success on tasks lacking paired data, including collection style transfer, object transfiguration, season transfer, and photo enhancement.
3
Introduced an unpaired image-to-image translation method that learns mapping G: X→Y without paired training examples using adversarial loss.
4
Provided quantitative comparisons showing the proposed approach outperforms several prior methods on unpaired translation tasks.
Research Object
Unpaired image-to-image translation between source domain X and target domain Y using cycle-consistent adversarial networks
Research Subject
Learning mappings G: X→Y and F:Y→X that produce target-distribution-consistent translated images while enforcing cycle-consistency (F(G(X)) ≈ X and G(F(Y)) ≈ Y) to enable image translation without paired training examples
Publication Details
Publication Date
2017-10-01
Journal
Publisher
ISSN
Cited by
24584
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai10
Deep Residual Learning for Image Recognition2016
Very Deep Convolutional Networks for Large-Scale Image Recognition2014
ImageNet: A large-scale hierarchical image database2009
Image-to-Image Translation with Conditional Adversarial Networks2017
Reducing the Dimensionality of Data with Neural Networks2006
Back-Translation for Cross-Cultural Research1970
Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network2017
Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks2015
Image Style Transfer Using Convolutional Neural Networks2016
Least Squares Generative Adversarial Networks2017
Cited by20
A survey on Image Data Augmentation for Deep Learning2019
Generative Adversarial Networks: An Overview2018
Self-Attention Generative Adversarial Networks2018
Deep Learning for Anomaly Detection: A Review2021
A Survey on Explainable Artificial Intelligence (XAI): Toward Medical XAI2020
Explainable Artificial Intelligence (XAI): What we know and what is left to attain Trustworthy Artificial Intelligence2023
Image Matching from Handcrafted to Deep Features: A Survey2020
Artificial intelligence in the creative industries: a review2021
Machine Learning and Deep Learning frameworks and libraries for large-scale data mining: a survey2019
Deep learning for AI2021
Deepfake Detection: A Systematic Literature Review2022
Artificial Intelligence and COVID-19: Deep Learning Approaches for Diagnosis and Treatment2020
Transparency of deep neural networks for medical image analysis: A review of interpretability methods2021
Artificial intelligence and machine learning for medical imaging: A technology review2021
Deep-Learning Inversion of Seismic Data2019
TadGAN: Time Series Anomaly Detection Using Generative Adversarial Networks2020
A survey on data‐efficient algorithms in big data era2021
Augmenting organizational decision-making with deep learning algorithms: Principles, promises, and challenges2020
Artificial Intelligence and Healthcare: A Journey through History, Present Innovations, and Future Possibilities2024
Generative AI-Driven Semantic Communication Networks: Architecture, Technologies, and Applications2024