Research map / 3D vision / Face and shape
TimeWalker: Personalized Neural Space for Lifelong Head Avatars
Dongwei Pan, Yang Li, Hongsheng LI, Kwan-Yee Lin
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 Dynamo instead of no Dynamo (ablation) | TimeWalker · Dynamo ablation (3 identities) | PSNR: 21.69 -> 27.20 | Improved |
| add Dynamo instead of a fixed single hashgrid (ablation) | TimeWalker · Dynamo ablation (3 identities) | PSNR: 24.84 -> 27.20 | Improved |
| use adaptive hashgrid number instead of all hashgrids (ablation) | TimeWalker · Dynamo ablation (3 identities) | PSNR: 26.86 -> 27.20 | Improved |
| use adaptive hashgrid number instead of 20 hashgrids (ablation) | TimeWalker · Dynamo ablation (3 identities) | PSNR: 26.94 -> 27.20 | Improved |
| use adaptive hashgrid number instead of 20 hashgrids (ablation) | TimeWalker · Dynamo ablation (3 identities) | LPIPS: 0.077 -> 0.077 | No clear effect |
| train on all lifestages instead of one stage (ablation) | TimeWalker · Multi- vs single-lifestage training | PSNR: 23.69 -> 27.22 | Improved |
| train on all lifestages instead of one stage (ablation) | TimeWalker · Multi- vs single-lifestage training | SSIM: 0.959 -> 0.955 | No clear effect |
| use TimeWalker instead of INSTA++ | Personalized head avatars · TimeWalker-1.0 Protocol-1 (1 vs 1) | PSNR: 26.39 -> 27.28 | Improved |
| use TimeWalker instead of Gaussian Surfels++ (one model per lifestage) | Personalized head avatars · TimeWalker-1.0 Protocol-2 (1 vs N) | PSNR: 27.61 -> 27.28 | Hurt |
| use TimeWalker instead of Gaussian Surfels++ (one model per lifestage) | Personalized head avatars · TimeWalker-1.0 Protocol-2 (1 vs N) | LPIPS: 0.134 -> 0.071 | Improved |
| add LPIPS loss (ablation) | TimeWalker · Loss ablation | rendering quality | Improved |
| add deformation regularization loss (ablation) | TimeWalker · Loss ablation | rendering quality | 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.
- FlashAvatar — FlashAvatar: High-fidelity Head Avatar with Efficient Gaussian Embedding (CVPR 2024) · better on 5 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
- Gaussian Eigen Models for Human Heads (CVPR 2025 · similarity 0.87)
- DiffusionAvatars: Deferred Diffusion for High-fidelity 3D Head Avatars (CVPR 2024 · similarity 0.86)
- CapHuman: Capture Your Moments in Parallel Universes (CVPR 2024 · similarity 0.86)
- Low-Rank Head Avatar Personalization with Registers (NeurIPS 2025 · similarity 0.86)
- FaceTalk: Audio-Driven Motion Diffusion for Neural Parametric Head Models (CVPR 2024 · similarity 0.85)
- Synthetic Prior for Few-Shot Drivable Head Avatar Inversion (CVPR 2025 · 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.
