Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | https://arxiv.org/abs/2605.30895 |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911738208714752 |
|---|---|
| author | Liu, Xuanyi Ji, Deyi Liu, Liqun Zhu, Lanyun Chen, Xuhang Xu, Qianxiong Shu, Peng Yu, Huan Jiang, Jie Gao, Feng Ma, Siwei |
| author_facet | Liu, Xuanyi Ji, Deyi Liu, Liqun Zhu, Lanyun Chen, Xuhang Xu, Qianxiong Shu, Peng Yu, Huan Jiang, Jie Gao, Feng Ma, Siwei |
| contents | Sparse camera-conditioned image-to-video generation presents a pivotal challenge: synthesizing geometrically consistent 3D motion from minimal pose cues. Existing methods, which largely rely on dense supervision or naive interpolation, suffer from severe pose drift and motion discontinuities due to the lack of robust 3D priors. In this paper, we introduce CamGeo, a novel framework that distills rich 3D geometric knowledge from a pre-trained video-to-3D model (VGGT) directly into the diffusion backbone. To achieve this without incurring inference latency, we propose a training-only distillation strategy. Specifically, CamGeo incorporates: (1) keyframe trajectory distillation that enforces cycle-consistency with sparse input poses, (2) cross-frame consistency distillation with both camera trajectory and depth constraints to generate consistent structure across unsupervised frames, and (3) a three-stage coarse-to-fine curriculum learning, progressively scales geometric complexity, from global structure coherence to fine-grained refinement, achieving stable optimization. Extensive experiments demonstrate that CamGeo achieves consistent improvements under various sparsity ratios. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_30895 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | CamGeo: Sparse Camera-Conditioned Image-to-Video Generation with 3D Geometry Priors Liu, Xuanyi Ji, Deyi Liu, Liqun Zhu, Lanyun Chen, Xuhang Xu, Qianxiong Shu, Peng Yu, Huan Jiang, Jie Gao, Feng Ma, Siwei Computational Engineering, Finance, and Science Sparse camera-conditioned image-to-video generation presents a pivotal challenge: synthesizing geometrically consistent 3D motion from minimal pose cues. Existing methods, which largely rely on dense supervision or naive interpolation, suffer from severe pose drift and motion discontinuities due to the lack of robust 3D priors. In this paper, we introduce CamGeo, a novel framework that distills rich 3D geometric knowledge from a pre-trained video-to-3D model (VGGT) directly into the diffusion backbone. To achieve this without incurring inference latency, we propose a training-only distillation strategy. Specifically, CamGeo incorporates: (1) keyframe trajectory distillation that enforces cycle-consistency with sparse input poses, (2) cross-frame consistency distillation with both camera trajectory and depth constraints to generate consistent structure across unsupervised frames, and (3) a three-stage coarse-to-fine curriculum learning, progressively scales geometric complexity, from global structure coherence to fine-grained refinement, achieving stable optimization. Extensive experiments demonstrate that CamGeo achieves consistent improvements under various sparsity ratios. |
| title | CamGeo: Sparse Camera-Conditioned Image-to-Video Generation with 3D Geometry Priors |
| topic | Computational Engineering, Finance, and Science |
| url | https://arxiv.org/abs/2605.30895 |