Research map / 3D vision / Gaussian splatting
VersatileGaussian: Real-time Neural Rendering for Versatile Tasks using Gaussian Splatting
Renjie Li, Zhiwen Fan, Bohua Wang, Peihao Wang, Zhangyang Wang, Xi Wu
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 feature fields (shared per-Gaussian features) instead of explicit labels (ablation) | 3D-GS multi-task · Replica | overall relative gain Delta_m (%): 0.00 -> +10.95 | Improved |
| add feature fields (shared per-Gaussian features) instead of explicit labels (ablation) | 3D-GS multi-task · Replica | RGB PSNR: 34.32 -> 32.86 | Hurt |
| add feature fields instead of explicit labels (ablation) | 3D-GS multi-task · ScanNet | RGB PSNR: 24.76 -> 24.60 | No clear effect |
| add Task Correlation Attention (TCA) (ablation) | Explicit + FF · Replica | RGB PSNR: 32.86 -> 34.57 | Improved |
| add Task Correlation Attention (TCA) (ablation) | Explicit + FF · ScanNet | overall relative gain Delta_m (%): +1.35 -> +8.65 | Improved |
| add TCA (ablation) | Explicit + FF · ScanNet | RGB PSNR: 24.60 -> 26.27 | Improved |
| replace NeRF-based multi-task view synthesis with Gaussian-splatting VersatileGaussian | SS-NeRF · Replica | FPS: 0.08 -> 34.68 | Improved |
| add keypoint and edge labels from off-the-shelf 2D methods to 3D-GS RGB training | 3D-GS · Tanks and Temples | RGB PSNR: 23.14 -> 25.57 | Improved |
| add only one extra task (keypoint) to 3D-GS | 3D-GS · Tanks and Temples | RGB PSNR: 23.14 -> 23.92 | Improved |
| use VersatileGaussian instead of invPT for semantic segmentation | invPT · ScanNet | semantic mIoU: 0.91 -> 0.90 | No clear effect |
Closest papers by meaning
- LightGaussian: Unbounded 3D Gaussian Compression with 15x Reduction and 200+ FPS (NeurIPS 2024 · similarity 0.89)
- HyRF: Hybrid Radiance Fields for Memory-efficient and High-quality Novel View Synthesis (NeurIPS 2025 · similarity 0.89)
- Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled Feature Fields (CVPR 2024 · similarity 0.89)
- Structure Consistent Gaussian Splatting with Matching Prior for Few-shot Novel View Synthesis (NeurIPS 2024 · similarity 0.89)
- MaskGaussian: Adaptive 3D Gaussian Representation from Probabilistic Masks (CVPR 2025 · similarity 0.88)
- Compact 3D Gaussian Representation for Radiance Field (CVPR 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.
