Research map / Label-efficient and robust learning / Domain adaptation and generalization
One-Shot Domain Adaptation for Face Generation
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 |
|---|---|---|---|
| Adapt StyleGAN to a one-shot target domain using manifold shifting and style mixing | StyleGAN-generated training data for face manipulation detection · Encoder-decoder Deepfake | Average Precision: 35.2% -> 93.4% | Helped |
| Adapt StyleGAN to a one-shot target domain using manifold shifting and style mixing | StyleGAN-generated training data for face manipulation detection · Face2Face | Average Precision: 35.3% -> 99.2 % | Helped |
| Adapt StyleGAN to a one-shot target domain using manifold shifting and style mixing | StyleGAN-generated training data for face manipulation detection · FaceSwap | Average Precision: 41.6% -> 95.2% | Helped |
| Add style mixing to manifold shifting (ablation) | StyleGAN one-shot adaptation · Encoder-decoder Deepfake | Average Precision: 43.1% -> 93.4% | Helped |
| Add manifold shifting to style mixing (ablation) | StyleGAN one-shot adaptation · Encoder-decoder Deepfake | Average Precision: 34.0% -> 93.4% | Helped |
| Train directly on one target example instead of using one-shot domain adaptation | Encoder-decoder Deepfake detection · Encoder-decoder Deepfake | Average Precision: 93.4 -> 52.1% | Hurt |
| Train directly on 10 target examples instead of using one-shot domain adaptation | Encoder-decoder Deepfake detection · Encoder-decoder Deepfake | Average Precision: 93.4 -> 79.7% | Hurt |
| Fine-tune StyleGAN with a single target example rather than use the proposed adaptation | StyleGAN for ProGAN-like image detection · ProGAN | Average Precision: 62.1% -> 10.2% | Hurt |
| Fine-tune StyleGAN with 10 target examples rather than use the proposed adaptation | StyleGAN for ProGAN-like image detection · ProGAN | Average Precision: 62.1% -> 21.8% | Hurt |
| Use a single reconstruction loss instead of combined L1 and perceptual loss (ablation) | Style-vector and generator-weight optimization · One-shot image reconstruction | reconstruction quality | Hurt |
| Update all StyleGAN layers during manifold shifting rather than a subset (ablation) | StyleGAN manifold shifting · Face manipulation detection | classification accuracy | Helped |
| Replace the final three style-vector layers with the target's layers (ablation) | StyleGAN style mixing · One-shot target image generation | visual similarity | Helped |
Closest papers by meaning
- Limitations of Face Image Generation (AAAI 2024 · similarity 0.86)
- DreamSalon: A Staged Diffusion Framework for Preserving Identity-Context in Editable Face Generation (CVPR 2024 · similarity 0.85)
- When StyleGAN Meets Stable Diffusion: a W+ Adapter for Personalized Image Generation (CVPR 2024 · similarity 0.85)
- CapHuman: Capture Your Moments in Parallel Universes (CVPR 2024 · similarity 0.85)
- Enhancing Identity-Deformation Disentanglement in StyleGAN for One-Shot Face Video Re-Enactment (AAAI 2025 · similarity 0.85)
- Link-based Contrastive Learning for One-Shot Unsupervised Domain Adaptation (CVPR 2025 · 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.