Research map / Point clouds / Point cloud classification
AvatarPointillist: AutoRegressive 4D Gaussian Avatarization
Hongyu Liu, Xuan Wang, Zijian Wu, Yating Wang, Ziyu Wan, Yue Ma, Runtao Liu, Boyao Zhou, Yujun Shen, Qifeng Chen
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 |
|---|---|---|---|
| Generate the Gaussian point cloud with the AR model instead of a fixed FLAME template (ablation) | AvatarPointillist Gaussian decoder · NeRSemble self-reenactment | FID: 120.34 -> 95.18 | Improved |
| Generate the Gaussian point cloud with the AR model instead of a fixed FLAME template (ablation) | AvatarPointillist Gaussian decoder · NeRSemble self-reenactment | LPIPS: 0.23 -> 0.15 | Improved |
| Add AR hidden features to the decoder input alongside positional embeddings (ablation) | AvatarPointillist Gaussian decoder · NeRSemble self-reenactment | FID: 103.80 -> 95.18 | Improved |
| Add AR hidden features to the decoder input alongside positional embeddings (ablation) | AvatarPointillist Gaussian decoder · NeRSemble self-reenactment | AKD: 5.81 -> 2.38 | Improved |
| Add positional encoding of generated points alongside AR features in the decoder input (ablation) | AvatarPointillist Gaussian decoder · NeRSemble self-reenactment | FID: 110.93 -> 95.18 | Improved |
| Add positional encoding of generated points alongside AR features in the decoder input (ablation) | AvatarPointillist Gaussian decoder · NeRSemble self-reenactment | AKD: 5.89 -> 2.38 | Improved |
| Use only AR features versus only positional embeddings as decoder input (ablation) | AvatarPointillist Gaussian decoder · NeRSemble self-reenactment | APD: 32.96 vs 41.49 | Mixed |
| Use AvatarPointillist instead of GAGAvatar | single-image 4D avatar generation · NeRSemble cross-reenactment | APD: 170.12 -> 153.13 (Portrait4Dv2 142.93 is lower) | Hurt |
| Use AvatarPointillist instead of the best baseline on identity preservation | single-image 4D avatar generation · NeRSemble cross-reenactment | CLIPScore: 0.71 -> 0.75 | Improved |
Closest papers by meaning
- 3DGS-Avatar: Animatable Avatars via Deformable 3D Gaussian Splatting (CVPR 2024 · similarity 0.88)
- GAvatar: Animatable 3D Gaussian Avatars with Implicit Mesh Learning (CVPR 2024 · similarity 0.88)
- GASP: Gaussian Avatars with Synthetic Priors (CVPR 2025 · similarity 0.88)
- GPAvatar: High-fidelity Head Avatars by Learning Efficient Gaussian Projections (CVPR 2025 · similarity 0.87)
- GaussianAvatars: Photorealistic Head Avatars with Rigged 3D Gaussians (CVPR 2024 · similarity 0.87)
- AvatarArtist: Open-Domain 4D Avatarization (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.
