taesungp/contrastive-unpaired-translation
Contrastive unpaired image-to-image translation, faster and lighter training than cyclegan (ECCV 2020, in PyTorch)
Employs patchwise contrastive learning combined with adversarial training to eliminate hand-crafted cycle-consistency losses, achieving 2x faster training and 50% memory reduction versus CycleGAN. Supports both standard unpaired translation and single-image translation where each domain consists of a single photograph. Built in PyTorch with configurable CUT and FastCUT variants, integrating with Visdom for real-time monitoring and pytorch-fid for FID evaluation.
2,491 stars. No commits in the last 6 months.
Stars
2,491
Forks
446
Language
Python
License
—
Category
Last pushed
Sep 05, 2023
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/taesungp/contrastive-unpaired-translation"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Related models
yunjey/domain-transfer-network
TensorFlow Implementation of Unsupervised Cross-Domain Image Generation
PaddlePaddle/PaddleGAN
PaddlePaddle GAN library, including lots of interesting applications like First-Order motion...
tohinz/ConSinGAN
PyTorch implementation of "Improved Techniques for Training Single-Image GANs" (WACV-21)
sagiebenaim/DistanceGAN
Pytorch implementation of "One-Sided Unsupervised Domain Mapping" NIPS 2017
ZPdesu/SEAN
SEAN: Image Synthesis with Semantic Region-Adaptive Normalization (CVPR 2020, Oral)