Research map / Image generation / Editing and style transfer
Unpaired Image-To-Image Translation Using Cycle-Consistent Adversarial Networks
Jun-Yan Zhu, Taesung Park, Phillip Isola, Alexei A. Efros
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use cycle-consistency with adversarial loss (CycleGAN) instead of pixel loss + GAN | unpaired translation · Maps to aerial photos (AMT) | % Turkers labeled real: 0.7% -> 26.8% | Improved |
| use CycleGAN instead of BiGAN/ALI | unpaired translation · Aerial photos to maps (AMT) | % Turkers labeled real: 1.9% -> 23.2% | Improved |
| use CycleGAN instead of CoGAN | unpaired translation · Cityscapes labels to photos | per-pixel accuracy: 0.40 -> 0.52 | Improved |
| use CycleGAN instead of CoGAN | unpaired translation · Cityscapes photos to labels | class IOU: 0.08 -> 0.16 | Improved |
| train with paired data (pix2pix) instead of unpaired CycleGAN | CycleGAN · Cityscapes labels to photos | per-pixel accuracy: 0.52 -> 0.71 | Improved |
| train with paired data (pix2pix) instead of unpaired CycleGAN | CycleGAN · Cityscapes photos to labels | class IOU: 0.16 -> 0.32 | Improved |
| add cycle-consistency loss to the GAN loss (ablation) | CycleGAN · Cityscapes labels to photos (qualitative) | image quality | Improved |
| add the adversarial (GAN) loss to the cycle loss (ablation) | CycleGAN · Cityscapes labels to photos (qualitative) | image quality | Improved |
| use cycle loss in both directions instead of forward only or backward only (ablation) | CycleGAN · Cityscapes labels to photos (qualitative) | training stability | Improved |
| add identity mapping loss L_identity (ablation) | CycleGAN · Monet paintings to photos | color composition preservation | Improved |
Closest papers by meaning
- Unpaired Image-to-Image Translation using Adversarial Consistency Loss (ECCV 2020 · similarity 0.92)
- Unsupervised Image-to-Image Translation with Stacked Cycle-Consistent Adversarial Networks (ECCV 2018 · similarity 0.91)
- Breaking the Cycle - Colleagues Are All You Need (CVPR 2020 · similarity 0.91)
- Benign Examples: Imperceptible Changes Can Enhance Image Translation Performance (AAAI 2020 · similarity 0.91)
- Adversarial Self-Defense for Cycle-Consistent GANs (NeurIPS 2019 · similarity 0.91)
- Few-Shot Unsupervised Image-to-Image Translation (ICCV 2019 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
