Research map / 3D vision / Gaussian splatting
TokenGS: Decoupling 3D Gaussian Prediction from Pixels with Learnable Tokens
Jiawei Ren, Michal Jan Tyszkiewicz, Jiahui Huang, Zan Gojcic
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 visibility loss (ablation) | TokenGS · RealEstate10K (4 views) | PSNR: 0.4 PSNR improvement | Improved |
| increase Gaussian tokens from 1024 to 4096 (finetune) (ablation) | TokenGS · RealEstate10K (2 views) | PSNR: 28.02 -> 28.41 | Improved |
| add token tuning (ablation) | TokenGS (4096 tok) · RealEstate10K (2 views) | PSNR: 28.41 -> 28.82 | Improved |
| add token tuning (ablation) | TokenGS · DL3DV 4 views | PSNR: 23.02 -> 23.57 | Improved |
| add token tuning (ablation) | TokenGS (zero-shot 6-view) · DL3DV 6 views | PSNR: 23.69 -> 24.51 | Improved |
| add token tuning (ablation) | TokenGS (zero-shot 2-view) · DL3DV 2 views | PSNR: 19.35 -> 19.32 | No clear effect |
| tune Gaussian parameters directly instead of tokens (ablation) | TokenGS · DL3DV (test-time) | PSNR and geometry | Mixed |
| extend context to up to 4x more views than trained (ablation) | TokenGS (4-view base) · DL3DV | PSNR | Improved |
| add more token-tuning steps (10 up to 250) (ablation) | TokenGS · DL3DV | PSNR | Improved |
| directly regress 3D means instead of per-pixel depth (vs GS-LRM) | TokenGS · RealEstate10K camera noise (2 views) | PSNR and LPIPS gap | Improved |
| replace PM-loss finetuned DepthSplat with TokenGS | DepthSplat+PM · DL3DV view extrapolation | PSNR: 20.77 -> 20.98 | Improved |
| use TokenGS instead of BTimer | BTimer · Kubric 4D | PSNR: 24.45 -> 24.84 | Improved |
Later papers that compared with this method
1 later accepted paper reports TokenGS as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstruction (NeurIPS 2026 · better on 0 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- DepthSplat — DepthSplat: Connecting Gaussian Splatting and Depth (CVPR 2025) · better on 6 of 7 shared results
- MVSplat — MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images (ECCV 2024) · better on 6 of 7 shared results
- GS-LRM — GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting (ECCV 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
- SfM-Free 3D Gaussian Splatting via Hierarchical Training (CVPR 2025 · similarity 0.90)
- LP-3DGS: Learning to Prune 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.90)
- FlexGS: Train Once, Deploy Everywhere with Many-in-One Flexible 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- COLMAP-Free 3D Gaussian Splatting (CVPR 2024 · similarity 0.89)
- DOGS: Distributed-Oriented Gaussian Splatting for Large-Scale 3D Reconstruction Via Gaussian Consensus (NeurIPS 2024 · similarity 0.89)
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.
