Research map / Image generation / Editing and style transfer
H-Adapter: Pose-Robust Hairstyle Transfer via Attention-Derived, Source-Aligned Hair Masks
Seulgi Jeong, Yunseong Cho, Sanghun Park
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 region-specific loss (H-Adapter instead of IP-Adapter, t8 mask) (ablation) | H-Adapter · CelebA-HQ pose-different subset | FID: 15.27 -> 12.47 | Improved |
| add region-specific loss (vs IP-Adapter t8) (ablation) | H-Adapter · CelebA-HQ pose-different subset | FID_CLIP: 8.83 -> 3.98 | Improved |
| add region-specific loss (vs IP-Adapter t8) (ablation) | H-Adapter · CelebA-HQ pose-different subset | CLIP-I: 0.639 -> 0.659 | Improved |
| add region-specific loss (vs IP-Adapter t8) (ablation) | H-Adapter · CelebA-HQ pose-different subset | SSIM: 0.803 -> 0.831 | Improved |
| add region-specific loss (vs IP-Adapter t8) (ablation) | H-Adapter · CelebA-HQ pose-different subset | PSNR: 21.56 -> 23.06 | Improved |
| add region-specific loss (vs IP-Adapter with t2 mask) (ablation) | H-Adapter · CelebA-HQ pose-different subset | FID_CLIP: 4.26 -> 3.98 | Improved |
| add region-specific loss (vs IP-Adapter with t2 mask) (ablation) | H-Adapter · CelebA-HQ pose-different subset | CLIP-I: 0.651 -> 0.659 | Improved |
| H-Adapter vs HairFastGAN | H-Adapter · CelebA-HQ pose-different subset | PSNR: 24.40 -> 23.06 | Hurt |
| H-Adapter vs HairFastGAN | H-Adapter · CelebA-HQ pose-different subset | FID: 12.78 -> 12.47 | Improved |
| H-Adapter vs Stable-Hair | H-Adapter · VLM-as-a-judge GPT-4o | HFS: 2.90 -> 3.11 | Improved |
| H-Adapter vs HairCLIPv2 | H-Adapter · VLM-as-a-judge GPT-4o | HFS: 2.10 -> 3.11 | Improved |
| H-Adapter vs Style-Your-Hair on artifact quality | H-Adapter · VLM-as-a-judge GPT-4o | AQS: 3.61 -> 3.73 | No clear effect |
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.
- Stable-Hair — Stable-Hair: Real-World Hair Transfer via Diffusion Model (AAAI 2025) · better on 4 of 4 shared results
- HairCLIPv2 — HairCLIP: Design Your Hair by Text and Reference Image (CVPR 2022) · better on 4 of 4 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
- What to Preserve and What to Transfer: Faithful, Identity-Preserving Diffusion-based Hairstyle Transfer (AAAI 2025 · similarity 0.92)
- Style Your Hair: Latent Optimization for Pose-Invariant Hairstyle Transfer via Local-Style-Aware Hair Alignment (ECCV 2022 · similarity 0.92)
- HairFastGAN: Realistic and Robust Hair Transfer with a Fast Encoder-Based Approach (NeurIPS 2024 · similarity 0.91)
- Stable-Hair: Real-World Hair Transfer via Diffusion Model (AAAI 2025 · similarity 0.90)
- HairDiffusion: Vivid Multi-Colored Hair Editing via Latent Diffusion (NeurIPS 2024 · similarity 0.86)
- MonoHair: High-Fidelity Hair Modeling from a Monocular Video (CVPR 2024 · similarity 0.86)
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.
