Research map / 3D vision / Face and shape
StyleGAN Salon: Multi-View Latent Optimization for Pose-Invariant Hairstyle Transfer
Sasikarn Khwanmuang, Pakkapon Phongthawee, Patsorn Sangkloy, Supasorn Suwajanakorn
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 two pose-specific guide images with shared latent information instead of a single-view guide (ablation) | StyleGAN Salon · Hairstyle transfer | hairstyle accuracy: 27.9% -> 23.4% | Improved |
| Add a W-space hallucination stage before W+ refinement (ablation) | StyleGAN Salon · Hairstyle transfer | visual quality: 27.9% -> 4.5% | Improved |
| Add latent sharing between the two pose views (ablation) | StyleGAN Salon · Hairstyle transfer | hairstyle accuracy: 27.9% -> 25.2% | Improved |
| Add RGB pre-masking to the masked perceptual loss (ablation) | StyleGAN Salon · Hairstyle transfer | boundary naturalness | Improved |
| Use randomized interpolation when sharing W-space latent information across views (ablation) | StyleGAN Salon · Hairstyle transfer | realism | Improved |
| Randomize the interpolation coefficient rather than fixing alpha to 0.5 (ablation) | StyleGAN Salon · Hairstyle transfer | cross-view consistency | Improved |
| Add the 32x32 global MSE loss (ablation) | StyleGAN Salon · Hairstyle transfer | overall appearance fidelity | Improved |
| Use StyleGAN Salon rather than the strongest baseline in the all-pairs user study | Hairstyle transfer on FFHQ · FFHQ user study (All) | user preference (%): 18.5 -> 56.2 | Improved |
| Use StyleGAN Salon rather than Barbershop for self-transfer reconstruction | Hair reconstruction · Hair reconstruction self-transfer | PSNR: 29.18 -> 27.84 | Hurt |
| Use StyleGAN Salon rather than StyleYourHair for face-shape preservation | Face shape preservation · FFHQ test pairs | RMSE: 14.37 -> 10.89 | Improved |
| Use StyleGAN Salon rather than LOHO for hairstyle-transfer FID | Hairstyle transfer · Hairstyle FID test set | FID: 20.72 -> 21.02 | Hurt |
Closest papers by meaning
- Style Your Hair: Latent Optimization for Pose-Invariant Hairstyle Transfer via Local-Style-Aware Hair Alignment (ECCV 2022 · similarity 0.93)
- HairNeRF: Geometry-Aware Image Synthesis for Hairstyle Transfer (ICCV 2023 · similarity 0.92)
- HairFastGAN: Realistic and Robust Hair Transfer with a Fast Encoder-Based Approach (NeurIPS 2024 · similarity 0.92)
- LOHO: Latent Optimization of Hairstyles via Orthogonalization (CVPR 2021 · similarity 0.92)
- What to Preserve and What to Transfer: Faithful, Identity-Preserving Diffusion-based Hairstyle Transfer (AAAI 2025 · similarity 0.90)
- Stable-Hair: Real-World Hair Transfer via Diffusion Model (AAAI 2025 · 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.
