Research map / Image generation / Editing and style transfer
Scaling-Up Disentanglement for Image Translation
Aviv Gabbay, Yedid Hoshen
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 correlated-attribute branch x^corr (ablation) | OverLORD · AFHQ | FID: 55.9 -> 16.5 | Improved |
| add correlated-attribute branch x^corr (ablation) | OverLORD · AFHQ | LPIPS diversity: 0 -> 0.51 | Improved |
| add correlated-attribute branch x^corr (ablation) | OverLORD · AFHQ | domain classification accuracy from u_uncorr (lower is better): 0.80 -> 0.33 | Improved |
| add adversarial loss in synthesis stage (ablation) | OverLORD · AFHQ | FID: 29.1 -> 16.5 | Improved |
| add adversarial loss in synthesis stage (ablation) | OverLORD · AFHQ | domain classification accuracy from u_uncorr (lower is better): 0.33 -> 0.33 | No clear effect |
| add adversarial loss in synthesis stage (ablation) | OverLORD · AFHQ | LPIPS diversity: 0.45 -> 0.51 | Improved |
| OverLORD instead of LORD | latent-optimization disentanglement · AFHQ | FID: 97.1 -> 16.5 | Improved |
| OverLORD instead of FUNIT | image translation · CelebA identity translation | identity similarity (Id): 0.24 -> 0.63 | Improved |
| model hair as correlated attribute with mask (Ours corr vs uncorr) | OverLORD · CelebA-HQ male-to-female (F2M) | FID: 54.0 -> 28.1 | Improved |
| model hair as correlated attribute with mask (Ours corr vs uncorr) | OverLORD · CelebA-HQ male-to-female (M2F) | FID: 42.9 -> 16.4 | Improved |
| amortized encoder instead of latent optimization in disentanglement stage (ablation) | OverLORD · disentanglement | correlation of labeled and unlabeled representations | Hurt |
| increase bottleneck weight lambda_b above 0.1 (ablation) | OverLORD · disentanglement | reconstruction / convergence: lambda_b > 0.1 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- StarGAN-v2 — StarGAN v2: Diverse Image Synthesis for Multiple Domains (CVPR 2020) · better on 4 of 5 shared results
- DRIT — DRIT++: Diverse Image-to-Image Translation via Disentangled Representations (IJCV 2020) · better on 1 of 1 shared result
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
- Multi-mapping Image-to-Image Translation via Learning Disentanglement (NeurIPS 2019 · similarity 0.91)
- Image-to-Image Translation With Disentangled Latent Vectors for Face Editing (TPAMI 2023 · similarity 0.90)
- Unsupervised Robust Disentangling of Latent Characteristics for Image Synthesis (ICCV 2019 · similarity 0.90)
- Multi-Attribute Transfer via Disentangled Representation (AAAI 2019 · similarity 0.90)
- Diverse Image-to-Image Translation via Disentangled Representations (ECCV 2018 · similarity 0.90)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 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.
