Research map / Video / Video generation
VMem: Consistent Interactive Video Scene Generation with Surfel-Indexed View Memory
Runjia Li, Philip Torr, Andrea Vedaldi, Tomas Jakab
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 |
|---|---|---|---|
| retrieve context views via surfel index instead of most recent K views (ablation) | VMem (K=4) · RealEstate10K cycle trajectories | PSNR: 7.52 -> 14.82 | Improved |
| retrieve context views via surfel index instead of camera distance (ablation) | VMem (K=17) · RealEstate10K cycle trajectories | PSNR: 15.72 -> 18.15 | Improved |
| retrieve context views via surfel index instead of field-of-view overlap (ablation) | VMem (K=4) · RealEstate10K cycle trajectories | PSNR: 13.11 -> 14.82 | Improved |
| surfel retrieval versus camera distance on rotation distance at K=17 (ablation) | VMem (K=17) · RealEstate10K cycle trajectories | R_dist: 0.821 -> 0.892 | Hurt |
| use a temporal window at K=17 for translation distance (ablation) | VMem (K=17) · RealEstate10K cycle trajectories | T_dist: 0.254 -> 0.165 | Improved |
| reduce context views from 17 to 4 with LoRA fine-tuned generator plus VMem (ablation) | VMem · inference speed on RTX 4090 | seconds per frame: 50 seconds -> 4.2 seconds | Improved |
| use K=4 fine-tuned VMem instead of K=17 on long-term RealEstate10K (ablation) | VMem · RealEstate10K long-term | PSNR: 14.09 -> 13.12 | Hurt |
| add VMem memory to SEVA on short-term benchmark | SEVA · RealEstate10K short-term | PSNR: 18.33 -> 18.33 (K=17) | No clear effect |
Closest papers by meaning
- WorldMem: Long-term Consistent World Simulation with Memory (NeurIPS 2025 · similarity 0.88)
- SynCamMaster: Synchronizing Multi-Camera Video Generation from Diverse Viewpoints (ICLR 2025 · similarity 0.87)
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.87)
- WonderWorld: Interactive 3D Scene Generation from a Single Image (CVPR 2025 · similarity 0.87)
- Video Perception Models for 3D Scene Synthesis (NeurIPS 2025 · similarity 0.86)
- 3D-Mem: 3D Scene Memory for Embodied Exploration and Reasoning (CVPR 2025 · 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.
