Research map / 3D vision / 3D generation
SceneTok: A Compressed, Diffusable Token Space for 3D Scenes
Mohammad Asim, Christopher Wewer, Jan Eric Lenssen
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 SceneTok compressed tokens instead of LVSM latent tokens | novel view synthesis · RE10K (12 context views) | rFVD: 211.66 -> 79.80 | Improved |
| use SceneTok instead of DepthSplat | novel view synthesis · RE10K (12 context views) | SSIM: 0.809 -> 0.783 | Hurt |
| use SceneTok instead of RayZer | novel view synthesis · DL3DV-140 | PSNR: 26.89 -> 21.95 | Hurt |
| use SceneTok instead of Long-LRM | novel view synthesis · DL3DV-140 | rFID: PSNR 23.06 -> 21.95; rFID 26.56 -> 19.12 | Mixed |
| use SceneTok instead of LVSM for rendering novel trajectories | transferability · DL3DV-140 scene pairs | rotation accuracy @10deg (%): 48.07 -> 75.81 | Improved |
| use only 2D RoPE instead of 3D RoPE in the encoder (ablation) | SceneTok encoder · transferability to novel trajectories | qualitative | Improved |
| use latent SceneGen instead of SEVA for single-view generation | scene generation · RE10K single-view generation | gFID: 17.69 -> 18.90 | Hurt |
| use latent SceneGen instead of DFoT | scene generation · RE10K single-view generation | inference time (s): 146 -> 26 | Improved |
| mask scene tokens or context views (reduce information) (ablation) | SceneTok decoder · RE10K qualitative variance | per-pixel output variance | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 4 on most shared results.
- DepthSplat — DepthSplat: Connecting Gaussian Splatting and Depth (CVPR 2025) · better on 2 of 5 shared results
- RayZer — RayZer: A Self-supervised Large View Synthesis Model (ICCV 2025) · better on 1 of 1 shared result
- MVSplat360 — MVSplat360: Feed-Forward 360 Scene Synthesis from Sparse Views (NeurIPS 2024) · better on 1 of 1 shared result
- MVSplat — MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images (ECCV 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
- Efficient Long Video Tokenization via Coordinate-based Patch Reconstruction (CVPR 2025 · similarity 0.87)
- UniFlow: A Unified Pixel Flow Tokenizer for Visual Understanding and Generation (ICLR 2026 · similarity 0.87)
- Good Token Hunting: A Hitchhiker's Guide to Token Selection for Visual Geometry Transformers (NeurIPS 2026 · similarity 0.87)
- FlexTok: Resampling Images into 1D Token Sequences of Flexible Length (ICML 2025 · similarity 0.87)
- When Worse is Better: Navigating the Compression Generation Trade-off In Visual Tokenization (NeurIPS 2025 · similarity 0.87)
- PPE: Positional Preservation Embedding for Token Compression in Multimodal Large Language Models (ICLR 2026 · similarity 0.87)
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.
