Research map / 3D vision / Gaussian splatting
Language-Driven Physics-Based Scene Synthesis and Editing via Feature Splatting
Ri-Zhao Qiu, Ge Yang, Weijia Zeng, Xiaolong Wang
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 SAM-based mask pooling and DINOv2 regularization to CLIP feature distillation (ablation) | Feature Splatting · LERF localization (2D) | localization accuracy: 73.0% -> 81.7% | Improved |
| add DINOv2 regularization (without SAM) on top of CLIP-only (ablation) | Feature Splatting · LERF localization (2D) | localization accuracy: 73.0% -> 71.4% | Hurt |
| add DINO regularization (ablation) | Feature Splatting · Collected datasets, validation views | CLIP feature loss (val): 0.048 -> 0.046 | Improved |
| add DINO regularization (ablation) | Feature Splatting · Collected datasets, training views | CLIP feature loss (train): 0.029 -> 0.032 | Hurt |
| add feature splatting to Gaussian splatting | Vanilla GS · Collected datasets | PSNR: 26.45 -> 26.47 | No clear effect |
| add feature splatting to Gaussian splatting | Vanilla GS · Collected datasets | FPS: 129 -> 102 | Hurt |
| add custom CUDA kernel optimizations (half2, shared gradient buffer, interleaved memory) (ablation) | Feature rasterizer, dim 768 · DB dataset, 1000 iterations | training time: 3.21 -> 1.21 | Improved |
| use low-dimensional (32) latent features with a shallow MLP decoder instead of 256 dims | Feature Splatting · DB dataset, 1000 iterations | training time: 0.41 -> 0.06 | Improved |
| add implicit volume preservation (densified surface and transparent internal particles) (ablation) | GS-Taichi-MPM · bouncing ball scene | qualitative physical plausibility | Improved |
| estimate Gaussian rotation from nearest-neighbor displacement instead of deformation gradient (ablation) | GS-Taichi-MPM · sculpture under large deformation | qualitative surface artifacts | Improved |
| use feature splatting instead of LERF for 2D localization | LERF baseline · LERF localization (2D) | localization accuracy: 80.3% -> 81.7% | Improved |
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.
- LERF — LERF: Language Embedded Radiance Fields (ICCV 2023) · 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
- SceneSplat++: A Large Dataset and Comprehensive Benchmark for Language Gaussian Splatting (NeurIPS 2025 · similarity 0.90)
- Spacetime Gaussian Feature Splatting for Real-Time Dynamic View Synthesis (CVPR 2024 · similarity 0.89)
- 3DitScene: Editing Any Scene via Language-guided Disentangled Gaussian Splatting (ICLR 2025 · similarity 0.89)
- Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled Feature Fields (CVPR 2024 · similarity 0.88)
- FHGS: Feature-Homogenized Gaussian Splatting (NeurIPS 2025 · similarity 0.88)
- Dr. Splat: Directly Referring 3D Gaussian Splatting via Direct Language Embedding Registration (CVPR 2025 · similarity 0.88)
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.
