Research map / Image generation / Editing and style transfer
CONFIG: Controllable Neural Face Image Generation
Marek Kowalski, Stephan J. Garbin, Virginia Estellers, Tadas Baltrušaitis, Matthew Johnson, Jamie Shotton
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 second training stage (real encoder and relaxed latent) before sampling z via latent GAN (ablation) | ConfigNet · FFHQ val (FID) | FID: 43.05 -> 39.76 | Improved |
| add synthetic data to training (ablation) | ConfigNet · FFHQ val (FID) | FID: 33.49 -> 33.41 | No clear effect |
| add the domain adversarial discriminator D_DA (ablation) | ConfigNet · CelebA attribute predictor on FFHQ val | C_pred(I+) - C_pred(I-): 0.20 -> 0.50 | Improved |
| add the first training stage (ablation) | ConfigNet · CelebA attribute predictor on FFHQ val | C_pred(I+) - C_pred(I-): 0.29 -> 0.50 | Improved |
| add the first training stage and D_DA, which slightly raise MD of non-altered attributes (ablation) | ConfigNet · CelebA attribute predictor on FFHQ val | MD: 0.03 -> 0.06 | Hurt |
| add one-shot fine-tuning on the single input image (ablation) | ConfigNet · CelebA attribute predictor on FFHQ val | C_pred(I+) - C_pred(I-): 0.50 -> 0.47 | No clear effect |
| use ConfigNet instead of CycleGAN on synthetic images for attribute control | attribute editing · Controllability vs CycleGAN (suppl.) | MD: 0.080 -> 0.055 | Improved |
| use ConfigNet instead of CycleGAN for photorealism | face generation · FFHQ val (FID, suppl.) | FID: 37.74 -> 33.41 | Improved |
| use CycleGAN instead of ConfigNet for dynamic range of attribute control | attribute editing · Controllability vs CycleGAN (suppl.) | C_pred(I+) - C_pred(I-): 0.50 -> 0.51 | No clear effect |
Closest papers by meaning
- High Resolution Zero-Shot Domain Adaptation of Synthetically Rendered Face Images (ECCV 2020 · similarity 0.89)
- Mixture of Global and Local Experts With Diffusion Transformer for Controllable Face Generation (TPAMI 2026 · similarity 0.88)
- CapHuman: Capture Your Moments in Parallel Universes (CVPR 2024 · similarity 0.86)
- FaceA-Net: Facial Attribute-Driven ID Preserving Image Generation Network (AAAI 2025 · similarity 0.86)
- ControlFace: Harnessing Facial Parametric Control for Face Rigging (CVPR 2025 · similarity 0.86)
- MagicPose: Realistic Human Poses and Facial Expressions Retargeting with Identity-aware Diffusion (ICML 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.
