Research map / Video / Video generation
GeoWorld: Providing Full-frame Geometry Features to Facilitate 3D Scene Generation
Yuhao Wan, Lijuan Liu, Jingzhi Zhou, Zihan Zhou, Xuying Zhang, dongbo zhang, Shaohui Jiao, Qibin Hou, Ming-Ming Cheng
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 geometric alignment loss (ablation) | GeoWorld geometry-constrained model · RealEstate10K | PSNR: 16.77 -> 16.96 | Improved |
| add geometric alignment loss (ablation) | GeoWorld geometry-constrained model · RealEstate10K | LPIPS: 0.3381 -> 0.3284 | Improved |
| add resize in geometry adaptation module (ablation) | GeoWorld · RealEstate10K | PSNR: 16.96 -> 17.17 | Improved |
| add global weighting with token discarding (ablation) | GeoWorld · RealEstate10K | PSNR: 17.17 -> 17.28 | Improved |
| add global weighting with token discarding (ablation) | GeoWorld · RealEstate10K | LPIPS: 0.3286 -> 0.3292 | No clear effect |
| discard 50% instead of 30% of low-weight tokens (ablation) | GeoWorld · RealEstate10K | PSNR: 17.17 -> 17.28 | Improved |
| discard 50% instead of 70% of low-weight tokens (ablation) | GeoWorld · RealEstate10K | PSNR: 17.22 -> 17.28 | Improved |
| use GeoWorld full-frame geometry instead of Hunyuan-Voyager | Hunyuan-Voyager · RealEstate10K novel view synthesis | PSNR: 14.85 -> 17.28 | Improved |
| use GeoWorld instead of FlexWorld | FlexWorld · RealEstate10K novel view synthesis | FID: 30.56 -> 31.00 | Hurt |
| use GeoWorld instead of FlexWorld | FlexWorld · RealEstate10K novel view synthesis | FVD: 270.4 -> 311.7 | Hurt |
| use GeoWorld instead of Hunyuan-Voyager for 3D scene generation | Hunyuan-Voyager · RealEstate10K 3D scene generation | PSNR: 13.52 -> 16.64 | Improved |
| two small video models with VGGT geometry instead of one large model | Hunyuan-Voyager · Cost analysis | inference time (s): 1110 -> 148 | Improved |
Closest papers by meaning
- Geometry-as-context: Modulating Explicit 3D in Scene-consistent Video Generation to Geometry Context (CVPR 2026 · similarity 0.90)
- SGAM: Building a Virtual 3D World through Simultaneous Generation and Mapping (NeurIPS 2022 · similarity 0.89)
- GeoVideo: Introducing Geometric Regularization into Video Generation Model (NeurIPS 2025 · similarity 0.89)
- WorldStereo: Bridging Camera-Guided Video Generation and Scene Reconstruction via 3D Geometric Memories (CVPR 2026 · similarity 0.89)
- WonderWorld: Interactive 3D Scene Generation from a Single Image (CVPR 2025 · similarity 0.88)
- PanoWorld: Geometry-Consistent Panoramic Video World Modeling (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.
