unpaired image-to-image translation with GANs
Torch implementation of CycleGAN for unpaired image-to-image translation between visual domains using cycle-consistent adversarial networks.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
Choose when you are working in PaddlePaddle and want a broader GAN toolbox with ready-to-use image translation and style-transfer applications instead of the original Torch CycleGAN code.
PaddlePaddle GAN library, including lots of interesting applications like First-Order motion transfer, Wav2Lip, picture repair, image editing, photo2cartoon, image style transfer, GPEN, and so on.
Choose when you need multiple diverse outputs for the same unpaired image-to-image translation task, at the cost of an older, less maintained codebase.
Multimodal Unsupervised Image-to-Image Translation
Choose when you want fast one-step image-to-image translation driven by Stable Diffusion with no per-dataset training and can accept the diffusion-model dependency.
One-step image-to-image with Stable Diffusion turbo: sketch2image, day2night, and more
Choose when training speed and memory matter and you still need unpaired translation; the official CUT code is faster and lighter than CycleGAN.
Contrastive unpaired image-to-image translation, faster and lighter training than cyclegan (ECCV 2020, in PyTorch)
Choose when no training data is available for the target translation and you want zero-shot text-prompt-guided image editing via Stable Diffusion.
Zero-shot Image-to-Image Translation [SIGGRAPH 2023]
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
Choose when you specifically need the older shared-latent-space UNIT baseline for unpaired translation; it is more limited and unmaintained compared with CycleGAN.
Unsupervised Image-to-Image Translation
Choose when your only need is portrait-to-cartoon conversion and you want a pretrained task-specific solution rather than training a generic CycleGAN.
人像卡通化探索项目 (photo-to-cartoon translation project)
These projects were analysed and named CycleGAN among their alternatives. The relationship is not symmetric — how CycleGAN rates them is a separate judgement, made when CycleGAN is analysed in its own right.
calls CycleGAN “when you need unpaired image-to-image translation without aligned training pairs.”
Image-to-image translation with conditional adversarial nets