Research map / Image generation / Editing and style transfer
Estimating the Success of Unsupervised Image to Image Translation
Sagie Benaim, Tomer Galanti, Lior Wolf
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 the proposed bound instead of GAN A loss as error predictor | DiscoGAN stopping criterion · Maps | Pearson correlation with ground truth error: -0.81 -> 1.00 | Improved |
| use the proposed bound instead of cycle A loss | DiscoGAN stopping criterion · Facades | Pearson correlation with ground truth error: 0.92 -> 0.85 | Mixed |
| use the proposed bound instead of cycle A loss | DiscoGAN stopping criterion · Shoes2Edges | Pearson correlation with ground truth error: 0.76 -> 1.00 | Improved |
| use the proposed bound instead of distance correlation loss | DistanceGAN stopping criterion · Bags2Edges | Pearson correlation with ground truth error: 0.34 -> 0.93 | Improved |
| use the proposed bound instead of distance correlation loss | DistanceGAN stopping criterion · Maps | Pearson correlation with ground truth error: 0.39 -> 1.00 | Improved |
| use the proposed bound instead of GAN losses for model selection | DiscoGAN number of layers · Cityscapes | Pearson correlation with ground truth error: 0.85 -> 0.99 | Improved |
| apply bound to per-sample error rather than expectation | DiscoGAN · Maps | Pearson correlation with ground truth error: 1.00 -> 0.66 | Mixed |
| apply bound to per-sample error for DistanceGAN | DistanceGAN · Cityscapes | Pearson correlation with ground truth error: 0.59 -> 0.51 | Mixed |
| tune hyperparameters with unsupervised hyperband using the bound instead of only number of layers (ablation) | DiscoGAN/DistanceGAN · five datasets | ground truth error of selected configuration | Improved |
| apply unsupervised hyperband to UNIT | UNIT · Maps | ground truth error of selected configuration | Improved |
Closest papers by meaning
- Risk Bounds for Unsupervised Cross-Domain Mapping with IPMs (JMLR 2021 · similarity 0.91)
- Few-Shot Unsupervised Image-to-Image Translation (ICCV 2019 · similarity 0.87)
- Unsupervised Image-to-Image Translation With Generative Prior (CVPR 2022 · similarity 0.87)
- Benign Examples: Imperceptible Changes Can Enhance Image Translation Performance (AAAI 2020 · similarity 0.87)
- Unsupervised Image-to-Image Translation Networks (NeurIPS 2017 · similarity 0.87)
- Towards Identifiable Unsupervised Domain Translation: A Diversified Distribution Matching Approach (ICLR 2024 · similarity 0.85)
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.
