Research map / 3D vision / Gaussian splatting
GaussianLens: Localized High-Resolution Reconstruction via On-Demand Gaussian Densification
Yijia Weng, Zhicheng Wang, Songyou Peng, Saining Xie, Howard Zhou, Leonidas Guibas
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 pixel-guided Gaussians alongside input Gaussians for densification (ablation) | GaussianLens · DL3DV 256->1024 | PSNR: 23.27 -> 23.62 (vs input only K=4) | Improved |
| increase densification factor K from 4 to 16 (ablation) | GaussianLens (input only) · DL3DV 256->1024 | PSNR: 23.27 -> 23.25 | No clear effect |
| add reconstruction-residual image features (ablation) | GaussianLens · DL3DV 256->1024 | PSNR: 23.19 -> 23.62 | Improved |
| add gradient features to Gaussian features (ablation) | GaussianLens · DL3DV 256->1024 | PSNR: 23.46 -> 23.62 | Improved |
| add multi-view features (ablation) | GaussianLens · DL3DV 256->1024 | PSNR: 23.41 -> 23.62 | Improved |
| add projection-based image cross-attention (ablation) | GaussianLens · DL3DV 256->1024 | PSNR: 22.59 -> 23.62 | Improved |
| use projection-based instead of global cross-attention (ablation) | GaussianLens · DL3DV 256->1024 | PSNR: 22.64 -> 23.62 | Improved |
| apply cross-attention at multiple scales instead of last block only (ablation) | GaussianLens · DL3DV 256->1024 | PSNR: 23.47 -> 23.62 | Improved |
| densify local RoI vs standard low-res reconstruction | GaussianLens · DL3DV 256->1024 | PSNR: 22.31 -> 23.62 | Improved |
| use existence-mask regularization to prune Gaussians (ablation) | GaussianLens · DL3DV 256->1024 | PSNR / Num. GS: 23.62 -> 23.56 PSNR; 220K -> 170K Gaussians | Mixed |
| apply trained model to unseen Gaussian source (zero-shot) | GaussianLens · RE10K / DL3DV | PSNR: per-scene optim. 22.34 -> 24.46 | Improved |
| crop high-res images instead of using local densification of initial Gaussians | DepthSplat high-res crop · DL3DV 256->1024 | PSNR: 22.31 -> 19.51 | Hurt |
Closest papers by meaning
- Generative Densification: Learning to Densify Gaussians for High-Fidelity Generalizable 3D Reconstruction (CVPR 2025 · similarity 0.91)
- CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstruction (NeurIPS 2026 · similarity 0.90)
- Perceptual-GS: Scene-adaptive Perceptual Densification for Gaussian Splatting (ICML 2025 · similarity 0.89)
- S2Gaussian: Sparse-View Super-Resolution 3D Gaussian Splatting (CVPR 2025 · similarity 0.89)
- HiSplat: Hierarchical 3D Gaussian Splatting for Generalizable Sparse-View Reconstruction (ICLR 2025 · similarity 0.89)
- No Pose, No Problem: Surprisingly Simple 3D Gaussian Splats from Sparse Unposed Images (ICLR 2025 · 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.
