Research map / Image generation / Editing and style transfer
Semantically Robust Unpaired Image Translation for Data With Unmatched Semantics Statistics
Zhiwei Jia, Bodi Yuan, Kangkang Wang, Hong Wu, David Clifford, Zhiqiang Yuan, Hao Su
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 |
|---|---|---|---|
| add semantic robustness loss | CUT · Cityscapes Label to Image | mIoU: 23.43 -> 27.23 | Improved |
| add semantic robustness loss | CUT · Cityscapes Label to Image | pixel accuracy: 75.09 -> 80.70 | Improved |
| add semantic robustness loss | CUT · GTA to Cityscapes | mIoU: 20.35 -> 22.69 | Improved |
| SRUNIT vs CycleGAN on mIoU | SRUNIT · GTA to Cityscapes | mIoU: 23.84 -> 22.69 | Hurt |
| add semantic robustness loss | CUT · Google Map to Aerial Photo | L2 distance: 70.28 -> 68.55 | Improved |
| add semantic robustness loss | CUT · Aerial Photo to Google Map | pixel accuracy (delta1): 16.25 -> 17.67 | Improved |
| add robustness loss rather than DistanceGAN self-distance constraint (ablation) | CUT · Cityscapes Label to Image | mIoU: 23.52 -> 27.23 | Improved |
| add robustness loss rather than HarmonicGAN-style loss (ablation) | CUT · Cityscapes Label to Image | mIoU: 23.86 -> 27.23 | Improved |
| add feature extractors F_k inside the robustness loss (ablation) | SRUNIT · Cityscapes Label to Image | mIoU: 23.89 -> 27.23 | Improved |
| use Eqn. 3 form rather than Eqn. 4 surrogate (ablation) | SRUNIT · Cityscapes Label to Image | mIoU: 24.71 -> 27.23 | Improved |
| perturb inputs rather than directly minimizing input-output semantics distance (ablation) | SRUNIT · Cityscapes Label to Image | mIoU: 23.22 -> 27.23 | Improved |
| constrain generator robustness rather than Lipschitz penalty on discriminator (ablation) | SRUNIT · Cityscapes Label to Image | mIoU: 23.51 -> 27.23 | Improved |
Closest papers by meaning
- Alleviating Semantics Distortion in Unsupervised Low-Level Image-to-Image Translation via Structure Consistency Constraint (CVPR 2022 · similarity 0.90)
- Unpaired Image-to-Image Translation using Adversarial Consistency Loss (ECCV 2020 · similarity 0.89)
- Unpaired Image-To-Image Translation Using Cycle-Consistent Adversarial Networks (ICCV 2017 · similarity 0.88)
- Contrastive Learning for Unpaired Image-to-Image Translation (ECCV 2020 · similarity 0.88)
- Unsupervised Image-to-Image Translation With Generative Prior (CVPR 2022 · similarity 0.88)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (AAAI 2024 · similarity 0.87)
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.
