Saved in:
| Main Authors: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2603.17117 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917351126990848 |
|---|---|
| author | Yu, Wei Qian, Runjia Li, Yumeng Wang, Liquan Yin, Songheng P, Sri Siddarth Chakaravarthy Anthony, Dennis Ye, Yang Li, Yidi Wan, Weiwei Garg, Animesh |
| author_facet | Yu, Wei Qian, Runjia Li, Yumeng Wang, Liquan Yin, Songheng P, Sri Siddarth Chakaravarthy Anthony, Dennis Ye, Yang Li, Yidi Wan, Weiwei Garg, Animesh |
| contents | Video diffusion models are moving beyond short, plausible clips toward world simulators that must remain consistent under camera motion, revisits, and intervention. Yet spatial memory remains a key bottleneck: explicit 3D structures can improve reprojection-based consistency but struggle to depict moving objects, while implicit memory often produces inaccurate camera motion even with correct poses. We propose Mosaic Memory (MosaicMem), a hybrid spatial memory that lifts patches into 3D for reliable localization and targeted retrieval, while exploiting the model's native conditioning to preserve prompt-following generation. MosaicMem composes spatially aligned patches in the queried view via a patch-and-compose interface, preserving what should persist while allowing the model to inpaint what should evolve. With PRoPE camera conditioning and two new memory alignment methods, experiments show improved pose adherence compared to implicit memory and stronger dynamic modeling than explicit baselines. MosaicMem further enables minute-level navigation, memory-based scene editing, and autoregressive rollout. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_17117 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | MosaicMem: Hybrid Spatial Memory for Controllable Video World Models Yu, Wei Qian, Runjia Li, Yumeng Wang, Liquan Yin, Songheng P, Sri Siddarth Chakaravarthy Anthony, Dennis Ye, Yang Li, Yidi Wan, Weiwei Garg, Animesh Computer Vision and Pattern Recognition Video diffusion models are moving beyond short, plausible clips toward world simulators that must remain consistent under camera motion, revisits, and intervention. Yet spatial memory remains a key bottleneck: explicit 3D structures can improve reprojection-based consistency but struggle to depict moving objects, while implicit memory often produces inaccurate camera motion even with correct poses. We propose Mosaic Memory (MosaicMem), a hybrid spatial memory that lifts patches into 3D for reliable localization and targeted retrieval, while exploiting the model's native conditioning to preserve prompt-following generation. MosaicMem composes spatially aligned patches in the queried view via a patch-and-compose interface, preserving what should persist while allowing the model to inpaint what should evolve. With PRoPE camera conditioning and two new memory alignment methods, experiments show improved pose adherence compared to implicit memory and stronger dynamic modeling than explicit baselines. MosaicMem further enables minute-level navigation, memory-based scene editing, and autoregressive rollout. |
| title | MosaicMem: Hybrid Spatial Memory for Controllable Video World Models |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.17117 |