Research map / Video / Video generation
CameraNoise: Enabling Faithful Camera Control in Video Diffusion through Geometry-Flow-Guided Noise Warping
Haoyu Zhao, Jiaxi Gu, Haoran Chen, Qingping Zheng, Yeying Jin, Hongyi Yang, Junqi Cheng, Yuang Zhang, Zenghui Lu, Huan Yu, Jie Jiang, Peng Shu and 2 more
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 dynamic scaling training (DST) of rotation scale with eta in (0.9,1.1) (ablation) | CameraNoise · RealEstate100 test (I2V) | FVD: 198.04 -> 180.81 | Improved |
| add dynamic scaling training (DST) of rotation scale with eta in (0.9,1.1) (ablation) | CameraNoise · RealEstate100 test (I2V) | TransErr: 0.167 -> 0.144 | Improved |
| sample eta in (0.9,1.1) instead of fixed eta=0.9 (ablation) | DST · RealEstate100 test (I2V) | FVD: 188.36 -> 180.81 | Improved |
| narrow eta range from (0.8,1.2) to (0.9,1.1) (ablation) | DST · RealEstate100 test (I2V) | FVD: 182.95 -> 180.81 | Improved |
| add GRFlow smoothing alpha=0.8 (vs alpha=1.0) (ablation) | GRFlow · RealEstate100 test (I2V) | FVD: +7.86% | Improved |
| set smoothing alpha=0.0 (vs alpha=1.0) (ablation) | GRFlow · RealEstate100 test (I2V) | TransErr: -376.19% | Hurt |
| set smoothing alpha=0.0 (vs alpha=1.0) (ablation) | GRFlow · RealEstate100 test (I2V) | FVD: -54.61% | Hurt |
| strengthen smoothing alpha=0.2 (vs alpha=1.0) (ablation) | GRFlow · RealEstate100 test (I2V) | Aesthetic Quality: -15.47% | Hurt |
| mix Gaussian noise into CameraNoise with lambda=0.6 (vs lambda=0.0) (ablation) | CameraNoise inference · RealEstate100 test (I2V) | TransErr: -45.15% | Hurt |
| mix Gaussian noise with lambda=1.0 (vs lambda=0.0) (ablation) | CameraNoise inference · RealEstate100 test (I2V) | RotErr: -121.16% | Hurt |
| mix Gaussian noise with lambda=1.0 (vs lambda=0.0) (ablation) | CameraNoise inference · RealEstate100 test (I2V) | Imaging Quality: +4.79% | Improved |
| use CameraNoise instead of best prior method | video diffusion · RealEstate100 test (I2V) | Dynamic Degree: +75.8% | Improved |
Closest papers by meaning
- Go-with-the-Flow: Motion-Controllable Video Diffusion Models Using Real-Time Warped Noise (CVPR 2025 · similarity 0.88)
- AC3D: Analyzing and Improving 3D Camera Control in Video Diffusion Transformers (CVPR 2025 · similarity 0.88)
- Training-free Camera Control for Video Generation (ICLR 2025 · similarity 0.88)
- VD3D: Taming Large Video Diffusion Transformers for 3D Camera Control (ICLR 2025 · similarity 0.87)
- Dynamic View Synthesis as an Inverse Problem (NeurIPS 2025 · similarity 0.87)
- ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models (CVPR 2024 · 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.
