Research map / Image generation / GANs
Instance-Wise Hard Negative Example Generation for Contrastive Learning in Unpaired Image-to-Image Translation
Weilun Wang, Wengang Zhou, Jianmin Bao, Dong Chen, Houqiang Li
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 negative generator with diversity loss (ablation) | CUT-style contrastive translation · Cityscapes | FID: 49.7 -> 48.5 | Improved |
| add negative generator with diversity loss (ablation) | CUT-style contrastive translation · Cat->Dog | FID: 110.9 -> 55.8 | Improved |
| add negative generator with diversity loss (ablation) | CUT-style contrastive translation · Horse->Zebra | FID: 59.6 -> 39.6 | Improved |
| add negative generator with diversity loss (ablation) | CUT-style contrastive translation · Cityscapes | pAcc: 71.9 -> 71.4 | Hurt |
| add diversity loss to negative generator (ablation) | NEGCUT · Cityscapes | FID: 91.5 -> 48.5 | Improved |
| add diversity loss to negative generator (ablation) | NEGCUT · Cat->Dog | FID: 83.0 -> 55.8 | Improved |
| add diversity loss to negative generator (ablation) | NEGCUT · Horse->Zebra | FID: 72.1 -> 39.6 | Improved |
| use 256 negatives instead of 64 (ablation) | NEGCUT · Cat->Dog | FID: 59.3 -> 55.8 | Mixed |
| use 256 negatives instead of 128 (ablation) | NEGCUT · Cat->Dog | FID: 86.9 -> 55.8 | Improved |
| use 256 negatives instead of 512 (ablation) | NEGCUT · Cat->Dog | FID: 62.8 -> 55.8 | Improved |
| use 256 negatives instead of 512 (ablation) | NEGCUT · Horse->Zebra | FID: 44.0 -> 39.6 | Improved |
| replace random negatives with generated hard negatives | CUT · Cityscapes | FID: 56.4 -> 48.5 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DRIT — DRIT++: Diverse Image-to-Image Translation via Disentangled Representations (IJCV 2020) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Contrastive Learning for Unpaired Image-to-Image Translation (ECCV 2020 · similarity 0.91)
- Unpaired Image-to-Image Translation using Adversarial Consistency Loss (ECCV 2020 · similarity 0.88)
- Unpaired Image-To-Image Translation Using Cycle-Consistent Adversarial Networks (ICCV 2017 · similarity 0.88)
- Semantically Robust Unpaired Image Translation for Data With Unmatched Semantics Statistics (ICCV 2021 · similarity 0.88)
- Few-Shot Unsupervised Image-to-Image Translation (ICCV 2019 · similarity 0.87)
- 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.
