Research map / Image generation / Editing and style transfer
StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 one target-conditioned generator across multiple domains instead of training separate pairwise models | StarGAN · CelebA single-attribute transfer: Hair color | AMT preference: 20.0% -> 66.2% | Helped |
| Use one target-conditioned generator across multiple domains instead of training separate pairwise models | StarGAN · CelebA single-attribute transfer: Gender | AMT preference: 31.4% -> 39.1% | Helped |
| Use one target-conditioned generator across multiple domains instead of training separate pairwise models | StarGAN · CelebA single-attribute transfer: Aged | AMT preference: 13.3% -> 70.6% | Helped |
| Train for flexible multi-attribute target labels rather than fixed pairwise translations | StarGAN · CelebA multi-attribute transfer: G+A | AMT preference: 20.3% -> 49.8% | Helped |
| Train with all available expression domains rather than only a source-target pair | StarGAN · RaFD facial expression synthesis | Classification error: 4.10 -> 2.12 | Helped |
| Use a mask vector to identify which dataset's labels are known | StarGAN-JNT · CelebA images with RaFD expression targets | Visual quality of facial expression synthesis | Helped |
| Jointly train on CelebA and RaFD instead of training only on RaFD (ablation) | StarGAN-JNT · CelebA images with RaFD expression targets | Visual quality of facial expression synthesis | Helped |
| Use the correct RaFD mask [0,1] instead of the wrong mask [1,0] (ablation) | StarGAN-JNT · CelebA images with RaFD expression targets | Expression synthesis quality | Helped |
Closest papers by meaning
- StarGAN v2: Diverse Image Synthesis for Multiple Domains (CVPR 2020 · similarity 0.92)
- TraVeLGAN: Image-To-Image Translation by Transformation Vector Learning (CVPR 2019 · similarity 0.90)
- Unsupervised Image-to-Image Translation Networks (NeurIPS 2017 · similarity 0.90)
- Conditional Image-to-Image Translation (CVPR 2018 · similarity 0.89)
- Unsupervised Image-to-Image Translation With Generative Prior (CVPR 2022 · similarity 0.89)
- Polymorphic-GAN: Generating Aligned Samples Across Multiple Domains With Learned Morph Maps (CVPR 2022 · similarity 0.89)
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.