Research map / Image generation / Editing and style transfer
Few-Shot Adversarial Learning of Realistic Neural Talking Head Models
Egor Zakharov, Aliaksandra Shysheya, Egor Burkov, Victor Lempitsky
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 |
|---|---|---|---|
| fine-tune the meta-learned model on a person's frames (FT) instead of feed-forward (FF) prediction | Ours · VoxCeleb2, T=32 | FID: 40.4 -> 30.6 | Improved |
| fine-tune instead of feed-forward at T=32 | Ours · VoxCeleb2, T=32 | CSIM: 0.48 -> 0.45 | Hurt |
| fine-tune instead of feed-forward at T=1 | Ours · VoxCeleb2, T=1 | FID: 46.1 -> 48.5 | Hurt |
| increase number of fine-tuning frames from 1 to 32 | Ours · VoxCeleb1 | FID: 43.0 -> 29.5 | Improved |
| use our model instead of X2Face | Ours · VoxCeleb1, T=32 | user study (USER): 0.85 -> 0.61 | Improved |
| use our model instead of Pix2pixHD | Ours · VoxCeleb1, T=32 | FID: 24.0 -> 29.5 | Hurt |
| use our model instead of Pix2pixHD | Ours · VoxCeleb1, T=32 | user study (USER): 0.71 -> 0.61 | Improved |
| train on larger VoxCeleb2 instead of VoxCeleb1 | Ours-FT · VoxCeleb1 vs VoxCeleb2, T=32 | FID, CSIM, USER: 29.5 -> 30.6 | Mixed |
| initialize generator and embedding randomly instead of from the embedder (ablation) | Ours · fine-tuning | image plausibility | Improved |
| add person-specific initialization of the discriminator (L_MCH) (ablation) | Ours · one-shot meta-learning | realism and personalization | Improved |
| add adversarial term during fine-tuning (ablation) | Ours · fine-tuning | image crispness | Improved |
Closest papers by meaning
- Speech2Face: Learning the Face Behind a Voice (CVPR 2019 · similarity 0.87)
- Free-HeadGAN: Neural Talking Head Synthesis With Explicit Gaze Control (TPAMI 2023 · similarity 0.86)
- Depth-Aware Generative Adversarial Network for Talking Head Video Generation (CVPR 2022 · similarity 0.86)
- 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 0.85)
- FACIAL: Synthesizing Dynamic Talking Face With Implicit Attribute Learning (ICCV 2021 · 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.
