Research map / 3D vision / Face and shape
Free-HeadGAN: Neural Talking Head Synthesis With Explicit Gaze Control
Michail Christos Doukas, Evangelos Ververas, Viktoriia Sharmanska, Stefanos P. Zafeiriou
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 canonical key-point estimator E_can (ablation) | Free-HeadGAN · VoxCeleb reenactment (15 pairs) | CSIM: 0.640 -> 0.789 | Improved |
| add canonical key-point estimator E_can (ablation) | Free-HeadGAN · VoxCeleb reenactment (15 pairs) | FID: 66.6 -> 53.9 | Improved |
| add canonical key-point estimator E_can (ablation) | Free-HeadGAN · VoxCeleb reenactment (15 pairs) | AU-H: 0.304 -> 0.351 | Hurt |
| add explicit gaze conditioning via E_gaze (ablation) | Free-HeadGAN · VoxCeleb reenactment (15 pairs) | AGD: 17.7 -> 13.3 | Improved |
| regress dense eye meshes instead of 3D gaze vectors (ablation) | E_gaze · Gaze360 | AGD: 12.7 -> 10.4 | Improved |
| regress dense eye meshes instead of 3D gaze vectors (ablation) | E_gaze · Columbia | AGD: 3.84 -> 1.12 | Improved |
| regress dense eye meshes instead of 3D gaze vectors (ablation) | E_gaze · MPIIGaze | AGD: 4.83 -> 3.35 | Improved |
| use 2 source images instead of 1 (ablation) | Free-HeadGAN · VoxCeleb self-reenactment (N-shot) | FID: 35.4 -> 32.7 | Improved |
| use 8 source images instead of 1 (ablation) | Free-HeadGAN · VoxCeleb self-reenactment (N-shot) | FID: 35.4 -> 29.9 | Improved |
| use 8 source images instead of 1 (ablation) | Free-HeadGAN · VoxCeleb self-reenactment (N-shot) | PSNR: 22.16 -> 24.96 | Improved |
| replace 3DMM with sparse 3D landmarks and canonical key-points | HeadGAN · VoxCeleb self-reenactment | FID: 36.1 -> 35.4 | Improved |
| replace 3DMM with sparse 3D landmarks and canonical key-points | HeadGAN · VoxCeleb reenactment (15 pairs) | CSIM: 0.688 -> 0.789 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HeadGAN — HeadGAN: One-Shot Neural Head Synthesis and Editing (ICCV 2021) · better on 10 of 11 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
- HeadGAN: One-Shot Neural Head Synthesis and Editing (ICCV 2021 · similarity 0.91)
- One-Shot High-Fidelity Talking-Head Synthesis With Deformable Neural Radiance Field (CVPR 2023 · similarity 0.89)
- FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models (CVPR 2024 · similarity 0.88)
- GoHD: Gaze-oriented and Highly Disentangled Portrait Animation with Rhythmic Poses and Realistic Expressions (AAAI 2025 · similarity 0.87)
- InsTaG: Learning Personalized 3D Talking Head from Few-Second Video (CVPR 2025 · similarity 0.87)
- SyncTalk: The Devil is in the Synchronization for Talking Head Synthesis (CVPR 2024 · similarity 0.87)
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.
