Research map / 3D vision / Gaussian splatting
Complexity-guided Regularization for Generalizable Human Gaussian Splatting
Gun Ryu, Gi-Mun Um, Won-Sik Cheong, Wonjun Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes complexity-guided regularization for generalizable human Gaussian splatting so that Gaussian density adapts to geometric and textural complexity of human subjects, to better capture fine details in novel-view synthesis of unseen humans.
Problem
Generalizable human Gaussian splatting methods struggle to represent fine-grained details of human subjects because they impose no constraint on how densely Gaussians are allocated relative to local geometric and textural complexity.
Approach
Regularizes Gaussian density according to the complexity of the human body region (geometric and textural), so that complex regions receive denser Gaussians; based only on the abstract, the specific mechanism is not described.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add complexity-guided regularization (ablation) | generalizable human Gaussian splatting · THuman2.0 | PSNR: 30.32 -> 30.79 | Improved |
| add complexity-aware patch-based supervision (ablation) | generalizable human Gaussian splatting · THuman2.0 | PSNR: 30.32 -> 32.06 | Improved |
| combine regularization and patch supervision (ablation) | generalizable human Gaussian splatting · THuman2.0 | PSNR: 32.06 -> 32.42 | Improved |
| add complexity-guided regularization (no depth loss) (ablation) | generalizable human Gaussian splatting · THuman2.0 (without depth regularization) | PSNR: 29.80 -> 30.41 | Improved |
| use complexity-aware patch sampling instead of random sampling (ablation) | patch-based supervision · THuman2.0 | PSNR: 31.63 -> 32.06 | Improved |
| use image-gradient sampling instead of random sampling (ablation) | patch-based supervision · THuman2.0 | PSNR: 31.63 -> 31.81 | Improved |
| use max of geometric and textural complexity instead of either alone (ablation) | complexity prior · THuman2.0 | PSNR: 31.65 -> 32.42 | Improved |
| use max instead of mean to combine complexities (ablation) | complexity prior · THuman2.0 | PSNR: 32.31 -> 32.42 | Improved |
| use pairwise ranking instead of Pearson correlation regularization (ablation) | complexity-guided regularization · THuman2.0 | PSNR: 30.31 -> 30.79 | Improved |
| increase nearest neighbors k from 7 to 9 (ablation) | density score · THuman2.0 | PSNR: 30.79 -> 30.75 | No clear effect |
| increase nearest neighbors k from 3 to 7 (ablation) | density score · THuman2.0 | PSNR: 30.55 -> 30.79 | Improved |
| add complexity-guided regularization (ablation) | Gaussian density · THuman2.0 | Spearman correlation: 0.1086 -> 0.4262 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- RoGSplat — RoGSplat: Learning Robust Generalizable Human Gaussian Splatting from Sparse Multi-View Images (CVPR 2025) · better on 4 of 4 shared results
- LIFe-GoM — LIFe-GoM: Generalizable Human Rendering with Learned Iterative Feedback Over Multi-Resolution Gaussians-on-Mesh (ICLR 2025) · better on 1 of 1 shared result
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
- HumanSplat: Generalizable Single-Image Human Gaussian Splatting with Structure Priors (NeurIPS 2024 · similarity 0.90)
- RoGSplat: Learning Robust Generalizable Human Gaussian Splatting from Sparse Multi-View Images (CVPR 2025 · similarity 0.89)
- Human Gaussian Splatting: Real-time Rendering of Animatable Avatars (CVPR 2024 · similarity 0.89)
- HUGS: Human Gaussian Splats (CVPR 2024 · similarity 0.88)
- GBC-Splat: Generalizable Gaussian-Based Clothed Human Digitalization under Sparse RGB Cameras (CVPR 2025 · similarity 0.88)
- FewViewGS: Gaussian Splatting with Few View Matching and Multi-stage Training (NeurIPS 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.
