Research map / 3D vision / Face and shape
DGTalker: Disentangled Generative Latent Space Learning for Audio-Driven Gaussian Talking Heads
Xiaoxi Liang, Yanbo Fan, Qiya Yang, Xuan Wang, Wei Gao, Ge Li
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 disentangled design (separate canonical, expression, lip codes) (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | PSNR: 27.741 -> 28.943 | Improved |
| add disentangled design (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | Sync: 3.869 -> 6.295 | Improved |
| add disentangled design (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | FID: 19.951 -> 15.149 | Improved |
| add dual encoders (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | Sync: 5.870 -> 6.295 | Improved |
| add dual encoders (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | PSNR: 28.473 -> 28.943 | Improved |
| add blendshapes (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | Sync: 6.189 -> 6.295 | Improved |
| add blendshapes (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | PSNR: 28.868 -> 28.943 | Improved |
| add masked cross-view supervision (MCS) (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | Sync: 4.547 -> 6.295 | Improved |
| add masked cross-view supervision (MCS) (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | LMD: 4.551 -> 3.997 | Improved |
| add masked cross-view supervision (MCS) (ablation) | DGTalker · Self-reconstruction ablation (8 clips) | PSNR: 28.559 -> 28.943 | Improved |
| use Gaussian generative prior with disentangled latent navigation instead of per-identity vanilla 3DGS | DGTalker vs TalkingGaussian · Novel-view self-reconstruction (5 viewpoints avg) | FID: 137.914 -> 80.011 | Improved |
| DGTalker vs TalkingGaussian inference speed | DGTalker · Self-reconstruction (8 clips) | FPS: 94 -> 71 | Hurt |
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.
- TalkingGaussian — TalkingGaussian: Structure-Persistent 3D Talking Head Synthesis via Gaussian Splatting (ECCV 2024) · better on 3 of 5 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
- MoDiTalker: Motion-Disentangled Diffusion Model for High-Fidelity Talking Head Generation (AAAI 2025 · similarity 0.90)
- PointTalk: Audio-Driven Dynamic Lip Point Cloud for 3D Gaussian-based Talking Head Synthesis (AAAI 2025 · similarity 0.89)
- GoHD: Gaze-oriented and Highly Disentangled Portrait Animation with Rhythmic Poses and Realistic Expressions (AAAI 2025 · similarity 0.88)
- Hierarchically Controlled Deformable 3D Gaussians for Talking Head Synthesis (AAAI 2025 · similarity 0.87)
- FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models (CVPR 2024 · similarity 0.87)
- Perceptually Accurate 3D Talking Head Generation: New Definitions, Speech-Mesh Representation, and Evaluation Metrics (CVPR 2025 · 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.
