Research map / 3D vision / Gaussian splatting
DLGStream: Dynamic Language-embedded Guassian Splatting for Open-vocabulary Enabled Free-viewpoint Video Streaming
ZHIHUI KE, Yvyang Liu, Xiaobo Zhou, Tie Qiu
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 separate opacity for color and language features (dual-opacity) (ablation) | DLGStream (3DGS) · N3DV 4D open-vocab query (dual-opacity ablation) | mIoU: 78.99 -> 85.78 | Improved |
| use separate opacity for color and language features (dual-opacity) (ablation) | DLGStream (3DGS) · N3DV 4D open-vocab query (dual-opacity ablation) | PSNR: 31.73 -> 32.15 | Improved |
| use separate opacity for color and language features (dual-opacity) (ablation) | DLGStream (Scaffold-GS) · N3DV 4D open-vocab query (dual-opacity ablation) | mIoU: 79.62 -> 84.69 | Improved |
| use bilinear instead of nearest interpolation of time features (ablation) | DLGStream (Scaffold-GS) · N3DV (interpolation method) | PSNR: 31.35 -> 31.93 | Improved |
| use bicubic instead of bilinear interpolation (ablation) | DLGStream (Scaffold-GS) · N3DV (interpolation method) | PSNR: 31.93 -> 32.20 | No clear effect |
| deform anchor attributes rather than decoded Gaussian attributes (ablation) | DLGStream (Scaffold-GS) · N3DV (Scaffold-GS deformation target) | PSNR: 29.37 -> 31.93 | Improved |
| enlarge voxel grids or MLP size for residual learning (ablation) | DLGStream (Scaffold-GS) · N3DV FVV reconstruction | PSNR: 31.92 - 31.94 | No clear effect |
| train GOPs in parallel from the first GOP (ablation) | DLGStream (3DGS) · N3DV parallel training | training time per frame (s): 54 -> 26 | Improved |
| skip frames in training and interpolate (3 FPS vs 30 FPS) (ablation) | DLGStream (3DGS) · N3DV frame interpolation | mIoU: 85.78 -> 82.64 | Mixed |
| GOP length 60 with key interval 10 vs other settings (ablation) | DLGStream (3DGS) · N3DV GOP/interval | PSNR: 32.26 at GOP 60/interval 10 | Mixed |
| replace 4DLangSplat with DLGStream on N3DV queries | 4DLangSplat · N3DV 4D open-vocab query | mIoU: 75.5 -> 85.8 | Improved |
| replace 4DLangSplat with DLGStream (unified language field) | 4DLangSplat · N3DV (language-embedded FVV) | storage (KB/frame): 431 -> 97.11 | 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.
- StreamSTGS — StreamSTGS: Streaming Spatial and Temporal Gaussian Grids for Real-Time Free-Viewpoint Video (AAAI 2026) · better on 2 of 3 shared results
- LangSplat — LangSplat: 3D Language Gaussian Splatting (CVPR 2024) · 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
- 3DGStream: On-the-Fly Training of 3D Gaussians for Efficient Streaming of Photo-Realistic Free-Viewpoint Videos (CVPR 2024 · similarity 0.90)
- Motion Matters: Compact Gaussian Streaming for Free-Viewpoint Video Reconstruction (NeurIPS 2025 · similarity 0.90)
- 4DGC: Rate-Aware 4D Gaussian Compression for Efficient Streamable Free-Viewpoint Video (CVPR 2025 · similarity 0.89)
- Compressing Streamable Free-Viewpoint Videos to 0.1 MB per Frame (AAAI 2025 · similarity 0.88)
- SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis (CVPR 2025 · similarity 0.88)
- QUEEN: QUantized Efficient ENcoding of Dynamic Gaussians for Streaming Free-viewpoint Videos (NeurIPS 2024 · 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.
