Research map / Image generation / Editing and style transfer
Retrieve in Style: Unsupervised Facial Feature Transfer and Retrieval
Min Jin Chong, Wen-Sheng Chu, Abhishek Kumar, David Forsyth
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 |
|---|---|---|---|
| compute contribution scores per image pair (N=2) instead of averaging over N=200 images (ablation) | RIS facial feature retrieval · CelebA-HQ retrieval, mouth | Attribute Matching Score (%): 75.8 -> 81.1 | Improved |
| compute contribution scores per image pair (N=2) instead of averaging over N=200 images (ablation) | RIS facial feature retrieval · CelebA-HQ retrieval, eyes | Attribute Matching Score (%): 95.4 -> 96.3 | Improved |
| compute contribution scores per image pair (N=2) instead of averaging over N=200 images (ablation) | RIS facial feature retrieval · CelebA-HQ retrieval, hair | Attribute Matching Score (%): 97.1 -> 97.5 | Improved |
| compute contribution scores per image pair (N=2) instead of averaging over N=200 images (ablation) | RIS facial feature retrieval · CelebA-HQ retrieval, nose | Attribute Matching Score (%): 100.0 -> 100.0 | No clear effect |
| use image-specific rather than averaged contribution scores (ablation) | RIS facial feature retrieval · TRSI-IoU, 100 queries | TRSI-IoU | Improved |
| switch from EIS to RIS for feature transfer | StyleGAN2 feature transfer · Image fidelity (CelebA-HQ transfers) | FID_inf: 3.47 -> 3.73 | Hurt |
| switch from EIS to RIS for feature transfer | StyleGAN2 feature transfer · Inference time (Titan Xp) | seconds per edit: 0.0394 -> 0.234 | Hurt |
| increase number of clusters K from 2 to 100 in submembership analysis (ablation) | contribution score channel overlap · Generated images (N=5000) | intersection ratio of top-100 channels | Hurt |
| vary random noise input (ablation) | RIS hair transfer · Generated images | absolute image difference | No clear effect |
Closest papers by meaning
- Face Image Retrieval With Attribute Manipulation (ICCV 2021 · similarity 0.88)
- StyleIPSB: Identity-Preserving Semantic Basis of StyleGAN for High Fidelity Face Swapping (CVPR 2023 · similarity 0.86)
- Aggregation via Separation: Boosting Facial Landmark Detector With Semi-Supervised Style Translation (ICCV 2019 · similarity 0.86)
- Fine-Grained Face Swapping via Regional GAN Inversion (CVPR 2023 · similarity 0.86)
- StyO: Stylize Your Face in Only One-Shot (AAAI 2025 · similarity 0.85)
- Content-Style Decoupling for Unsupervised Makeup Transfer without Generating Pseudo Ground Truth (CVPR 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.
