Research map / Image generation / Editing and style transfer
Bridging the Gap Between Label- and Reference-Based Synthesis in Multi-Attribute Image-to-Image Translation
Qiusheng Huang, Zhilin Zheng, Xueqi Hu, Li Sun, Qingli 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 the reference encoding module and latent cycle loss to the label-only model (ablation) | Model A · CelebA label-based | FID: 25.10 -> 17.07 | Improved |
| add the reference encoding module and latent cycle loss to the label-only model (ablation) | Model A · CelebA label-based | Accuracy: 84.3 -> 83.5 | No clear effect |
| add source encoding S_s to LEM and REM (ablation) | Model B · CelebA label-based | FID: 17.07 -> 12.14 | Improved |
| add source encoding S_s to LEM and REM (ablation) | Model B · CelebA reference-based | FID: 13.91 -> 9.52 | Improved |
| add interpolated image to adversarial loss (ablation) | Model C · CelebA label-based | Accuracy: 82.8 -> 86.2 | Improved |
| add interpolated image to adversarial loss (ablation) | Model C · CelebA label-based | FID: 12.14 -> 17.32 | Hurt |
| add mode seeking loss (ablation) | Model D · CelebA label-based | LPIPS: 0.012 -> 0.030 | Improved |
| add mode seeking loss (ablation) | Model D · CelebA label-based | FID: 17.32 -> 14.03 | Improved |
| add style closeness loss L_sty (ablation) | Model E · CelebA label-based | Accuracy: 82.9 -> 85.1 | Improved |
| add style closeness loss L_sty (ablation) | Model E · CelebA label-based | LPIPS: 0.030 -> 0.034 | Improved |
| add attribute keeping constraint L_ak (ablation) | Model F · CelebA label-based | Accuracy: 85.1 -> 87.0 | Improved |
| add attribute keeping constraint L_ak (ablation) | Model F · CelebA reference-based | Accuracy: 39.5 -> 54.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- StarGAN-V2 — StarGAN v2: Diverse Image Synthesis for Multiple Domains (CVPR 2020) · better on 3 of 3 shared results
- StarGAN — StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation (CVPR 2018) · better on 2 of 2 shared results
- ELEGANT — ELEGANT: Exchanging Latent Encodings with GAN for Transferring Multiple Face Attributes (ECCV 2018) · better on 2 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
- Smoothing the Disentangled Latent Style Space for Unsupervised Image-to-Image Translation (CVPR 2021 · similarity 0.88)
- Domain Adaptive Image-to-Image Translation (CVPR 2020 · similarity 0.87)
- LANIT: Language-Driven Image-to-Image Translation for Unlabeled Data (CVPR 2023 · similarity 0.87)
- Image-to-Image Translation via Hierarchical Style Disentanglement (CVPR 2021 · similarity 0.87)
- Continuous and Diverse Image-to-Image Translation via Signed Attribute Vectors (IJCV 2022 · similarity 0.87)
- Rethinking the Paradigm of Content Constraints in Unpaired Image-to-Image Translation (AAAI 2024 · similarity 0.85)
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.
