Research map / Image generation / Editing and style transfer
StarGAN v2: Diverse Image Synthesis for Multiple Domains
Check 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 StarGAN v2 instead of the previous leading methods for latent-guided synthesis | Image translation on CelebA-HQ and AFHQ · CelebA-HQ and AFHQ latent-guided synthesis | FID: 13.7 and 16.2; more than two times improvement over the previous leading method | Helped |
| Use StarGAN v2 instead of the competing methods | Image translation on CelebA-HQ · CelebA-HQ latent-guided synthesis | LPIPS | Helped |
| Use StarGAN v2 instead of the previous leading methods for reference-guided synthesis | Image translation on CelebA-HQ and AFHQ · CelebA-HQ and AFHQ reference-guided synthesis | FID: 23.8 and 19.8; about 1.5× and 3.5× better than the previous leading method | Helped |
| Use StarGAN v2 instead of the competing methods | Image translation on CelebA-HQ and AFHQ · CelebA-HQ and AFHQ reference-guided synthesis | LPIPS | Helped |
| Replace the ACGAN discriminator with a multi-task discriminator (ablation) | StarGAN baseline · CelebA-HQ | visual quality | Helped |
| Add R1 regularization and replace depth-wise concatenation with AdaIN (ablation) | StarGAN baseline with a multi-task discriminator · CelebA-HQ | training stability | Helped |
| Provide a latent code directly to the generator with latent reconstruction loss, rather than learning domain-specific style codes (ablation) | StarGAN multi-domain image translation · CelebA-HQ | image diversity | Hurt |
| Add a mapping network that transforms latent codes into domain-specific style codes, with style reconstruction loss (ablation) | StarGAN multi-domain image translation · CelebA-HQ | image diversity | Helped |
| Add diversity regularization (ablation) | StarGAN v2 without diversity regularization · CelebA-HQ | image diversity | Helped |
| Apply pixel normalization to the mapping network's latent code (ablation) | StarGAN v2 mapping network · Image translation tasks in the paper | model performance | No clear effect |
| Apply feature normalization to the mapping network (ablation) | StarGAN v2 mapping network · Image translation tasks in the paper | model performance | Hurt |
| Use feature normalization or PatchGAN in the discriminator (ablation) | StarGAN v2 discriminator · Image translation tasks in the paper | output quality | No clear effect |
Closest papers by meaning
- StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation (CVPR 2018 · similarity 0.92)
- TraVeLGAN: Image-To-Image Translation by Transformation Vector Learning (CVPR 2019 · similarity 0.87)
- DeepI2I: Enabling Deep Hierarchical Image-to-Image Translation by Transferring from GANs (NeurIPS 2020 · similarity 0.86)
- SynthRGB-T: Language-Vision Guided Image Translation for Diversity Synthesis (CVPR 2026 · similarity 0.85)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (AAAI 2024 · similarity 0.84)
- KOALA: Empirical Lessons Toward Memory-Efficient and Fast Diffusion Models for Text-to-Image Synthesis (NeurIPS 2024 · similarity 0.84)
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.