Research map / 3D vision / Gaussian splatting
CATSplat: Context-Aware Transformer with Spatial Guidance for Generalizable 3D Gaussian Splatting from A Single-View Image
Wonseok Roh, Hwanhee Jung, Jong Wook Kim, Seunggwan Lee, Innfarn Yoo, Andreas Lugmayr, Seunggeun Chi, Karthik Ramani, Sangpil Kim
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 contextual prior from VLM text embeddings via cross-attention (ablation) | CATSplat baseline transformer · RE10K (n=Random) | PSNR: 25.02 -> 25.40 | Improved |
| add spatial prior from 3D point features via cross-attention (ablation) | CATSplat baseline transformer · RE10K (n=Random) | PSNR: 25.02 -> 25.42 | Improved |
| combine both priors (ablation) | CATSplat · RE10K (n=Random) | PSNR: 25.02 -> 25.45 | Improved |
| use pretrained visual encoder features (DINOv2, ConvNeXt) instead of the proposed priors (ablation) | CATSplat cross-attention · RE10K (n=Random) | PSNR: 25.40 -> 25.17 (DINOv2-g vs contextual) | Hurt |
| use single-sentence text description rather than scene type or object list (ablation) | contextual prior · RE10K (n=Random) | PSNR: 25.13 -> 25.40 | Improved |
| extend text description to two or more sentences (ablation) | contextual prior · RE10K (n=Random) | PSNR: 25.40 -> 25.29 | Hurt |
| concatenate depth with image (ablation) | baseline transformer · RE10K (n=Random) | PSNR: 24.82 -> 25.02 | Improved |
| replace depth concatenation with unprojected 3D point concatenation (ablation) | baseline transformer · RE10K (n=Random) | PSNR: 25.02 -> 25.04 | No clear effect |
| cross-attend point features rather than depth features (ablation) | spatial prior · RE10K (n=Random) | PSNR: 25.16 -> 25.42 | Improved |
| increase number of cross-attention layers from one to three (ablation) | CATSplat · RE10K qualitative | image error maps | Improved |
| use CATSplat instead of Flash3D for extrapolation | single-view 3DGS · RE10K extrapolation | PSNR: 24.10 -> 25.35 | Improved |
| single-view CATSplat versus two-view MVSplat on interpolation | RE10K interpolation · RE10K interpolation | PSNR: 26.39 vs 25.23 | Hurt |
Closest papers by meaning
- pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction (CVPR 2024 · similarity 0.91)
- HiSplat: Hierarchical 3D Gaussian Splatting for Generalizable Sparse-View Reconstruction (ICLR 2025 · similarity 0.90)
- PF3plat: Pose-Free Feed-Forward 3D Gaussian Splatting for Novel View Synthesis (ICML 2025 · similarity 0.90)
- DeSplat: Decomposed Gaussian Splatting for Distractor-Free Rendering (CVPR 2025 · similarity 0.89)
- TranSplat: Generalizable 3D Gaussian Splatting from Sparse Multi-View Images with Transformers (AAAI 2025 · similarity 0.89)
- SceneSplat++: A Large Dataset and Comprehensive Benchmark for Language Gaussian Splatting (NeurIPS 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.
