Verified project record

junyanz/CycleGAN

CycleGAN learns image-to-image translation between two domains using unpaired training data, operating on Linux and OSX with an NVIDIA GPU. It provides pre-trained model inference, custom model training via Lua scripts, and a display server for monitoring training progress.

Project overview

Project type: Image & Vision · Infrastructure
Use cases: Image Processing
Deployment: Refer to project documentation
License: License pending

Best for

Researchers and developers working on Linux or OSX who need to translate images between two distinct visual domains without paired data.

Key capabilities

Translates images between domains using CycleGAN and pix2pix without requiring input-output pairs.
Allows users to download and run pre-trained models to generate images in various styles.
Enables training new CycleGAN models on user-provided datasets via train.lua.
Provides a web server to view training results as the model learns.

Limitations and risks

A lingering gap remains between the results achievable with paired training data and this unpaired method.
The method does not work well on random unrelated combinations like cats and keyboards.
An NVIDIA GPU with CUDA CuDNN is required, and CPU mode is untested.

Getting started

Setup involves manual installation of Torch and its dependencies. Users must clone the repository, download datasets and pre-trained models via shell scripts, and execute tests via Lua commands in the CLI.

Evidence and sources

GitHub project description: Software that can generate photos from paintings, turn horses into zebras, perform style transfer, and more.
README: Torch implementation for learning an image-to-image translation (i.e. [pix2pix](https://github.com/phillipi/pix2pix)) **without** input-output pairs, for example:
README: - NVIDIA GPU + CUDA CuDNN (CPU mode and CUDA without CuDNN may work with minimal modification, but untested)
README: - Install torch and dependencies from https://github.com/torch/distro - Install torch packages `nngraph`, `class`, `display` ```bash luarocks install nngraph luarocks install clas…
README: - Now, let's generate Paul Cézanne style images: ``` DATA_ROOT=./datasets/ae_photos name=style_cezanne_pretrained model=one_direction_test phase=test loadSize=256 fineSize=256 res…

Reviewed at: 2026-07-18T00:29:22Z