Research map / Video / Video generation
Spatia: Video Generation with Updatable Spatial Memory
Jinjing Zhao, Fangyun Wei, Zhening Liu, Hongyang Zhang, Chang Xu, Yan Lu
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 scene projection video conditioning (without reference frames) (ablation) | Spatia · WorldScore subset closed-loop | PSNR_C: 15.55 -> 17.18 | Improved |
| add scene projection video conditioning (ablation) | Spatia · WorldScore subset closed-loop | Camera control: 58.81 -> 80.13 | Improved |
| add reference frames on top of scene video (ablation) | Spatia · WorldScore subset closed-loop | PSNR_C: 17.18 -> 19.38 | Improved |
| add reference frames without scene video (ablation) | Spatia · WorldScore subset closed-loop | PSNR_C: 15.55 -> 15.64 | No clear effect |
| add both scene video and reference frames (ablation) | Spatia · WorldScore subset closed-loop | LPIPS_C: 0.379 -> 0.213 | Improved |
| increase number of reference frames from 1 to 7 (ablation) | Spatia · WorldScore subset closed-loop | PSNR_C: 17.50 -> 19.38 | Improved |
| increase number of reference frames from 1 to 7 (ablation) | Spatia · WorldScore subset closed-loop | Match accuracy: 0.592 -> 0.698 | Improved |
| add spatial memory (Spatia vs Wan2.2 baseline) at 6 clips | Wan2.2 (5B) · Long-horizon closed-loop (6 clips) | PSNR_C: 10.74 -> 18.04 | Improved |
| increase clips from 2 to 6 with spatial memory | Spatia · Long-horizon closed-loop | PSNR_C: 19.38 -> 18.04 | Hurt |
| use finer point cloud voxels (0.07 m to 0.01 m) (ablation) | Spatia · RealEstate test | PSNR: 15.97 -> 18.58 | Improved |
| Spatia vs Voyager in closed-loop memory evaluation | Voyager · WorldScore subset closed-loop | Match accuracy: 0.507 -> 0.698 | Improved |
Closest papers by meaning
- 3D Scene Prompting for Scene-Consistent Camera-Controllable Video Generation (ICLR 2026 · similarity 0.90)
- StarGen: A Spatiotemporal Autoregression Framework with Video Diffusion Model for Scalable and Controllable Scene Generation (CVPR 2025 · similarity 0.87)
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.86)
- GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control (CVPR 2025 · similarity 0.85)
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · similarity 0.85)
- Track4Gen: Teaching Video Diffusion Models to Track Points Improves Video Generation (CVPR 2025 · similarity 0.84)
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.
