Research map / Image generation / Editing and style transfer
Learned Spatial Representations for Few-Shot Talking-Head Synthesis
Moustafa Meshry, Saksham Suri, Larry S. Davis, Abhinav Shrivastava
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 |
|---|---|---|---|
| replace UNet generator with SPADE conditioned on landmarks (ablation) | FSTH-style landmark-driven generator · VoxCeleb ablation | FID: 67.19 -> 56.00 | Improved |
| replace UNet generator with SPADE conditioned on landmarks (ablation) | FSTH-style landmark-driven generator · VoxCeleb ablation | LPIPS: 0.274 -> 0.268 | No clear effect |
| add learned segmentation maps as SPADE input (ablation) | FSTH-style landmark-driven generator + SPADE · VoxCeleb ablation | identity similarity: 0.834 -> 0.828 | Hurt |
| add learned segmentation maps as SPADE input (ablation) | FSTH-style landmark-driven generator + SPADE · VoxCeleb ablation | FID: 56.00 -> 62.78 | Hurt |
| add learned segmentation maps as SPADE input (ablation) | FSTH-style landmark-driven generator + SPADE · VoxCeleb ablation | LPIPS: 0.268 -> 0.265 | Improved |
| replace supervised segmentation with learned latent spatial layout (ablation) | FSTH-style landmark-driven generator + learned seg. maps · VoxCeleb ablation | LPIPS: 0.265 -> 0.247 | Improved |
| replace supervised segmentation with learned latent spatial layout (ablation) | FSTH-style landmark-driven generator + learned seg. maps · VoxCeleb ablation | FID: 62.78 -> 54.40 | Improved |
| replace supervised segmentation with learned latent spatial layout (ablation) | FSTH-style landmark-driven generator + learned seg. maps · VoxCeleb ablation | identity similarity: 0.828 -> 0.860 | Improved |
| use oracle ground-truth segmentations as spatial input (upper bound) (ablation) | SPADE generator · VoxCeleb ablation | FID: 54.40 -> 48.06 | Improved |
| use proposed method instead of FOMM | talking-head synthesis · VoxCeleb single-shot | PSNR: 18.20 -> 17.37 | Hurt |
| use proposed method instead of FOMM | talking-head synthesis · VoxCeleb single-shot | identity similarity: 0.869 -> 0.886 | Improved |
| use proposed method instead of FSTH | talking-head synthesis · VoxCeleb single-shot | FID: 53.07 -> 45.69 | Improved |
Closest papers by meaning
- One-Shot High-Fidelity Talking-Head Synthesis With Deformable Neural Radiance Field (CVPR 2023 · similarity 0.88)
- MetaPortrait: Identity-Preserving Talking Head Generation With Fast Personalized Adaptation (CVPR 2023 · similarity 0.88)
- EDTalk: Efficient Disentanglement for Emotional Talking Head Synthesis (ECCV 2024 · similarity 0.88)
- FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models (CVPR 2024 · similarity 0.87)
- InsTaG: Learning Personalized 3D Talking Head from Few-Second Video (CVPR 2025 · similarity 0.86)
- AE-NeRF: Audio Enhanced Neural Radiance Field for Few Shot Talking Head Synthesis (AAAI 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.
