Research map / Image generation / Editing and style transfer
Domain-Scalable Unpaired Image Translation via Latent Space Anchoring
Siyu Huang, Jie An, Donglai Wei, Zudi Lin, Jiebo Luo, Hanspeter Pfister
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 256x256 feature maps instead of 4x4 as regressor input (ablation) | Latent Anchor · seg-to-RGB, StyleGAN2 FFHQ | FID: 201.5 -> 83.8 | Improved |
| use 256x256 feature maps instead of 16x16 (ablation) | Latent Anchor · seg-to-RGB, StyleGAN2 FFHQ | FID: 188.9 -> 83.8 | Improved |
| use 256x256 feature maps instead of 64x64 (ablation) | Latent Anchor · seg-to-RGB, StyleGAN2 FFHQ | FID: 115.6 -> 83.8 | Improved |
| use 256x256 feature maps instead of 128x128 (ablation) | Latent Anchor · seg-to-RGB, StyleGAN2 FFHQ | FID: 116.2 -> 83.8 | Improved |
| use only 256x256 feature map rather than concatenating all feature maps (ablation) | Latent Anchor · seg-to-RGB, StyleGAN2 FFHQ | FID: 129.5 -> 83.8 | Improved |
| use 6 conv layers regressor instead of 3 conv layers (ablation) | Latent Anchor · seg-to-RGB, StyleGAN2 FFHQ | FID: 93.8 -> 83.8 | Improved |
| use 6 conv layers regressor instead of 6 ResBlocks (ablation) | Latent Anchor · seg-to-RGB, StyleGAN2 FFHQ | FID: 155.1 -> 83.8 | Improved |
| add reconstruction loss (ablation) | Latent Anchor · seg-to-RGB face | qualitative reconstruction | Improved |
| add latent regularization loss (ablation) | Latent Anchor · seg-to-RGB face | qualitative image quality | Improved |
| add adversarial loss (ablation) | Latent Anchor · seg-to-RGB face | qualitative fidelity | Improved |
| pretrain the GAN generator on the input or output domain instead of another domain (ablation) | Latent Anchor · AFHQ-wild to AFHQ-cat | qualitative detail | No clear effect |
| Latent Anchor instead of CUT on segmentation-to-face | Latent Anchor · seg-to-face (fixed-domain UNIT) | FID: 38.9 -> 83.8 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- StarGANv2 — StarGAN v2: Diverse Image Synthesis for Multiple Domains (CVPR 2020) · better on 1 of 2 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
- Homomorphic Latent Space Interpolation for Unpaired Image-To-Image Translation (CVPR 2019 · similarity 0.92)
- Unsupervised Image-to-Image Translation With Generative Prior (CVPR 2022 · similarity 0.91)
- Exploring Explicit Domain Supervision for Latent Space Disentanglement in Unpaired Image-to-Image Translation (TPAMI 2019 · similarity 0.91)
- GP-UNIT: Generative Prior for Versatile Unsupervised Image-to-Image Translation (TPAMI 2023 · similarity 0.91)
- Homomorphic Interpolation Network for Unpaired Image-to-image Translation (TPAMI 2020 · similarity 0.91)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (AAAI 2024 · similarity 0.88)
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.
