Research map / 3D vision / Gaussian splatting
Generalizable Human Gaussians for Sparse View Synthesis
YoungJoong Kwon, Baole Fang, Yixing Lu, Haoye Dong, Cheng Zhang, Francisco Vicente Carrasco, Albert Mosella-Montoro, Jianjin Xu, Shingo J Takagi, Daeil Kim, Aayush Prakash, Fernando de la Torre
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 |
|---|---|---|---|
| use multiple scaffolds for geometry input, appearance input and Gaussian output instead of a single scaffold (ablation) | GHG · THuman (3 views) | FID: 84.38 -> 61.67 | Improved |
| add multi-scaffold Gaussian output maps instead of one large-scale or offset Gaussian map (ablation) | GHG · THuman (3 views) | FID: 90.16 -> 61.67 | Improved |
| add multi-scaffold geometry and appearance input (ablation) | GHG · THuman (3 views) | PSNR: 23.11 -> 21.90 (single large scaffold vs full) | Hurt |
| add geometry cue encoder (ablation) | GHG · THuman (3 views) | FID: 69.12 -> 61.67 | Improved |
| add appearance cue encoder (ablation) | GHG · THuman (3 views) | FID: 78.56 -> 61.67 | Improved |
| add 2D UV texture inpainting network (ablation) | GHG · THuman (3 views) | FID: 67.15 -> 61.67 | Improved |
| increase outer scaffolds from 0 to 4 (ablation) | GHG · THuman (3 views) | FID: 84.38 -> 61.67 | Improved |
| increase outer scaffolds beyond 4 (5 and 6) (ablation) | GHG · THuman (3 views) | FID: 61.67 -> 63.80 (5), 64.81 (6) | No clear effect |
| add SSIM, mask and multi-view losses to L1 (ablation) | GHG · THuman (3 views) | FID: 71.97 -> 61.67 | Improved |
| increase number of input views at inference from 3 to 5 (ablation) | GHG · THuman | FID: 61.67 -> 35.00 | Improved |
| replace per-subject vanilla Gaussians with GHG | Gaussian splatting for humans · THuman (5 views) | PSNR: 17.62 -> 22.06 | Improved |
| compare GHG with GPS-Gaussian on LPIPS | GHG · THuman (5 views) | LPIPS: 123.30 -> 132.42 | 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.
- GPS-Gaussian — GPS-Gaussian: Generalizable Pixel-wise 3D Gaussian Splatting for Real-time Human Novel View Synthesis (CVPR 2024) · better on 2 of 3 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
- GPS-Gaussian: Generalizable Pixel-wise 3D Gaussian Splatting for Real-time Human Novel View Synthesis (CVPR 2024 · similarity 0.93)
- FewViewGS: Gaussian Splatting with Few View Matching and Multi-stage Training (NeurIPS 2024 · similarity 0.92)
- Generative Sparse-View Gaussian Splatting (CVPR 2025 · similarity 0.92)
- Human Gaussian Splatting: Real-time Rendering of Animatable Avatars (CVPR 2024 · similarity 0.92)
- Generalizable Human Gaussians from Single-View Image (ICLR 2025 · similarity 0.92)
- HUGS: Human Gaussian Splats (CVPR 2024 · similarity 0.91)
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.
