Research map / 3D vision / Gaussian splatting
tttLRM: Test-Time Training for Long Context and Autoregressive 3D Reconstruction
Chen Wang, Hao Tan, Wang Yifan, Zhiqin Chen, Yuheng Liu, Kalyan Sunkavalli, Sai Bi, Lingjie Liu, Yiwei Hu
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 LVSM pretraining (GS head) (ablation) | tttLRM · GSO 256x256 | PSNR: 32.77 -> 33.14 | Improved |
| add LVSM pretraining (triplane) (ablation) | tttLRM · Triplane NeRF | PSNR: 26.40 -> 27.87 | Improved |
| re-predict full Gaussians each step instead of Predict & Merge (ablation) | tttLRM AR · DL3DV 32 views | PSNR: 21.50 -> 23.63 | Improved |
| use Muon optimizer (ablation) | tttLRM · 32 view 256x144 | PSNR: 20.44 -> 20.68 | Improved |
| add opacity and depth regularization (ablation) | tttLRM · 32 view 256x144 | PSNR: 20.68 -> 20.76 | Improved |
| tttLRM vs GS-LRM | tttLRM · GSO 512x512 24 views | PSNR: 33.26 -> 34.80 | Improved |
| tttLRM vs Long-LRM | tttLRM · DL3DV-140 32 views | PSNR: 24.10 -> 25.07 | Improved |
| tttLRM vs optimization baselines at 64 views | tttLRM · DL3DV-140 64 views | PSNR: 27.07 -> 25.95 | Hurt |
| autoregressive (causal) variant vs full feedforward | tttLRM · DL3DV-140 32 views | PSNR: 25.07 -> 24.31 | Hurt |
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.
- Scaffold-GS — Scaffold-GS: Structured 3D Gaussians for View-Adaptive Rendering (CVPR 2024) · better on 4 of 6 shared results
- GS-LRM — GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting (ECCV 2024) · better on 4 of 4 shared results
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
- TTT3R: 3D Reconstruction as Test-Time Training (ICLR 2026 · similarity 0.90)
- iLRM: An Iterative Large 3D Reconstruction Model (CVPR 2026 · similarity 0.88)
- DGS-LRM: Real-Time Deformable 3D Gaussian Reconstruction From Monocular Videos (NeurIPS 2025 · similarity 0.87)
- MegaSynth: Scaling Up 3D Scene Reconstruction with Synthesized Data (CVPR 2025 · similarity 0.87)
- pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction (CVPR 2024 · similarity 0.86)
- LRM: Large Reconstruction Model for Single Image to 3D (ICLR 2024 · similarity 0.86)
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.
