Research map / Video / Video generation
VGGT-World: Transforming VGGT into an Autoregressive Geometry World Model
Xiangyu Sun, Shijie Wang, Fengyi Zhang, Lin Liu, Caiyan Jia, Ziying Song, Zi Helen Huang, Yadan Luo
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 |
|---|---|---|---|
| predict clean latent (z-prediction) instead of velocity in 1024-d VGGT feature space (ablation) | VGGT-World flow transformer · VGGT layer-4 latent training | denoising quality / SNR | Improved |
| add Stage-2 trajectory-consistent flow forcing (ablation) | VGGT-World · Cityscapes mid-term | AbsRel: 0.131 -> 0.126 | Improved |
| add Stage-2 trajectory-consistent flow forcing (ablation) | VGGT-World · Cityscapes mid-term | delta1: 86.8 -> 87.3 | Improved |
| add Stage-2 trajectory-consistent flow forcing (ablation) | VGGT-World · TartanAir 1-view | Chamfer Distance: 3.0818 -> 2.8451 | Improved |
| use linear lambda schedule 0 to 1 instead of static lambda=0.1 (ablation) | VGGT-World flow forcing · Cityscapes mid-term | AbsRel: 0.142 -> 0.126 | Improved |
| use larger static mixing ratio (0.1 -> 0.3 -> 0.7) (ablation) | VGGT-World flow forcing · Cityscapes mid-term | AbsRel: 0.142 -> 0.148 -> 0.165 | Hurt |
| geometry latent state (VGGT) instead of pixel-space video generation (Cosmos) | world model for depth forecasting · KITTI short-term mean | AbsRel: 0.154 -> 0.065 | Improved |
| forecast in frozen VGGT feature space instead of DINO feature space (DINO-Foresight) | latent world model · KITTI mid-term mean | AbsRel: 0.115 -> 0.098 | Improved |
| use geometry latent forecasting instead of Gen3R video generation, pose-free | point cloud forecasting · TartanAir 1-view | Chamfer Distance: 5.7478 -> 2.8451 | Improved |
| compare against Gen3R given text conditioning | VGGT-World (pose-free) · TartanAir 1-view | Completeness/CD: Comp. 4.5297 vs 2.5367; CD 2.8451 vs 2.7809 | Mixed |
| forecast future horizon and jointly decode with the present frame (ablation) | VGGT-World · TartanAir short-horizon | Chamfer Distance | Improved |
| extend forecast horizon beyond about 7 frames (ablation) | VGGT-World joint decoding · TartanAir | Chamfer Distance | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Aether — Aether: Geometric-Aware Unified World Modeling (ICCV 2025) · better on 2 of 2 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
- 4D-VGGT: A SpatioTemporal Foundation Model for Dynamic Scene Geometry Estimation (ECCV 2026 · similarity 0.89)
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.88)
- MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting (NeurIPS 2026 · similarity 0.87)
- LoopNav: Benchmarking Spatial Consistency in World Models (NeurIPS 2026 · similarity 0.87)
- VGGT: Visual Geometry Grounded Transformer (CVPR 2025 · similarity 0.87)
- DriveDreamer-Policy: A Geometry-Grounded World-Action Model for Unified Generation and Planning (NeurIPS 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.
