# junyanz/cyclegan

**Attribution required: if you use, quote, or summarise this content, you must credit and link back to [awesome-repositories.com](https://awesome-repositories.com/repository/junyanz-cyclegan).**

12,861 stars · 1,957 forks · Lua · NOASSERTION

## Links

- GitHub: https://github.com/junyanz/CycleGAN
- awesome-repositories: https://awesome-repositories.com/repository/junyanz-cyclegan.md

## Description

CycleGAN is a generative adversarial network framework designed for unpaired image-to-image translation. It enables the conversion of images between two distinct visual domains using datasets that do not require direct one-to-one matching examples.

The project implements a deep learning style transfer tool capable of artistic style transfer, object transfiguration, and domain-to-domain conversion. It uses a dual-generator architecture and cycle-consistency loss to ensure that images translated to a target domain and back recover their original state.

The framework covers core machine learning workflows including generative translation model training and training data refinement to translate synthetic datasets into realistic styles. It also includes tools for real-time training visualization to monitor image transformations during the training and testing processes.

This project is built using PyTorch.

## Tags

### Artificial Intelligence & ML

- [Cycle Consistency Constraints](https://awesome-repositories.com/f/artificial-intelligence-ml/cycle-consistency-constraints.md) — Uses cycle-consistency loss to ensure images translated between domains can be reconstructed to their original form.
- [CycleGAN Training Utilities](https://awesome-repositories.com/f/artificial-intelligence-ml/cyclegan-training-utilities.md) — Provides utilities for training CycleGAN models on unpaired image datasets.
- [Dual-Generator Architectures](https://awesome-repositories.com/f/artificial-intelligence-ml/dual-generator-architectures.md) — Utilizes two separate generative networks to enable bidirectional image translation between distinct visual styles.
- [Generative Adversarial Networks](https://awesome-repositories.com/f/artificial-intelligence-ml/generative-adversarial-networks.md) — Implements a generative adversarial network architecture to synthesize realistic image translations.
- [Generative Model Training Tools](https://awesome-repositories.com/f/artificial-intelligence-ml/generative-model-training-tools.md) — Implements a training process for generative networks to enable conversion between two visual domains. ([source](https://github.com/junyanz/cyclegan#readme))
- [Neural Style Transfer](https://awesome-repositories.com/f/artificial-intelligence-ml/neural-style-transfer.md) — Utilizes deep learning to apply artistic characteristics and object transfigurations to photographs. ([source](https://junyanz.github.io/CycleGAN/))
- [PyTorch Training Frameworks](https://awesome-repositories.com/f/artificial-intelligence-ml/pytorch-training-frameworks.md) — Provides a PyTorch-based implementation for training and executing cyclic image translation models.
- [Unpaired Image Translation](https://awesome-repositories.com/f/artificial-intelligence-ml/unpaired-image-translation.md) — Enables image translation between domains using datasets that lack direct one-to-one matching examples.
- [Domain-Specific Discriminators](https://awesome-repositories.com/f/artificial-intelligence-ml/domain-specific-discriminators.md) — Employs separate classifier networks for each target domain to verify the stylistic accuracy of generated images.
- [Stylistic Data Refinement](https://awesome-repositories.com/f/artificial-intelligence-ml/machine-learning-data-augmentation/stylistic-data-refinement.md) — Refines synthetic training data by translating it into realistic styles for improved real-world performance. ([source](https://junyanz.github.io/CycleGAN/))
- [Object Class Transformations](https://awesome-repositories.com/f/artificial-intelligence-ml/object-class-transformations.md) — Changes the visual appearance of objects in an image to represent different categories or species.
- [Object Transfigurations](https://awesome-repositories.com/f/artificial-intelligence-ml/object-transfigurations.md) — Provides the capability to change the visual appearance of specific objects to represent different categories. ([source](https://junyanz.github.io/CycleGAN/))
- [Synthetic Data Refinement](https://awesome-repositories.com/f/artificial-intelligence-ml/synthetic-data-refinement.md) — Translates synthetic datasets into realistic styles to improve machine learning performance on real-world data.

### Part of an Awesome List

- [Image Translation Frameworks](https://awesome-repositories.com/f/awesome-lists/ai/image-translation-frameworks.md) — Implements a comprehensive framework for image-to-image translation without paired training examples.
- [Domain Transfer and Translation](https://awesome-repositories.com/f/awesome-lists/ai/domain-transfer-and-translation.md) — Unpaired image-to-image translation using cycle-consistency constraints.