Contrastive Unpaired Translation (CUT) is a PyTorch-based framework for unpaired image-to-image translation. It leverages patchwise contrastive learning and adversarial training to achieve high-quality translations without the need for paired datasets. CUT offers faster and more memory-efficient training compared to traditional methods like CycleGAN.
Key Features
Utilizes patchwise contrastive learning to align corresponding patches between input and output images.
Eliminates the need for hand-crafted loss functions and inverse networks.
Supports single-image training scenarios, enabling translation with minimal data.
Faster and more memory-efficient training compared to CycleGAN.
Based on PyTorch and compatible with existing CycleGAN and pix2pix frameworks.
Includes implementations of both CUT and FastCUT models.
Android DevHub
.png)
.png)
.png)
.png)