GEN3C-Testing-Example

# GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control
CVPR 2025 (Highlight)

[Xuanchi Ren*](https://xuanchiren.com/),
[Tianchang Shen*](https://www.cs.toronto.edu/~shenti11/)
[Jiahui Huang](https://huangjh-pub.github.io/),
[Huan Ling](https://www.cs.toronto.edu/~linghuan/),
[Yifan Lu](https://yifanlu0227.github.io/),
[Merlin Nimier-David](https://merlin.nimierdavid.fr/),
[Thomas Müller](https://research.nvidia.com/person/thomas-muller),
[Alexander Keller](https://research.nvidia.com/person/alex-keller),
[Sanja Fidler](https://www.cs.toronto.edu/~fidler/),
[Jun Gao](https://www.cs.toronto.edu/~jungao/) <br>
\* indicates equal contribution <br>
**[Paper](https://arxiv.org/pdf/2503.03751), [Project Page](https://research.nvidia.com/labs/toronto-ai/GEN3C/)**

Abstract: We present GEN3C, a generative video model with precise Camera Control and
temporal 3D Consistency. Prior video models already generate realistic videos,
but they tend to leverage little 3D information, leading to inconsistencies,
such as objects popping in and out of existence. Camera control, if implemented
at all, is imprecise, because camera parameters are mere inputs to the neural
network which must then infer how the video depends on the camera. In contrast,
GEN3C is guided by a 3D cache: point clouds obtained by predicting the
pixel-wise depth of seed images or previously generated frames. When generating
the next frames, GEN3C is conditioned on the 2D renderings of the 3D cache with
the new camera trajectory provided by the user. Crucially, this means that
GEN3C neither has to remember what it previously generated nor does it have to
infer the image structure from the camera pose. The model, instead, can focus
all its generative power on previously unobserved regions, as well as advancing
the scene state to the next frame. Our results demonstrate more precise camera
control than prior work, as well as state-of-the-art results in sparse-view
novel view synthesis, even in challenging settings such as driving scenes and
monocular dynamic video. Results are best viewed in videos.

## Citation
```
@inproceedings{ren2025gen3c,
title={GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control},
author={Ren, Xuanchi and Shen, Tianchang and Huang, Jiahui and Ling, Huan and
Lu, Yifan and Nimier-David, Merlin and Müller, Thomas and Keller, Alexander and
Fidler, Sanja and Gao, Jun},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
year={2025}
}

# GEN3C：融合三维信息的全局一致视频生成与精准相机控制
CVPR 2025（高亮点论文）

[任轩驰*](https://xuanchiren.com/), [沈天畅*](https://www.cs.toronto.edu/~shenti11/), [黄佳辉](https://huangjh-pub.github.io/), [凌欢](https://www.cs.toronto.edu/~linghuan/), [陆一帆](https://yifanlu0227.github.io/), [Merlin Nimier-David](https://merlin.nimierdavid.fr/), [Thomas Müller](https://research.nvidia.com/person/thomas-muller), [Alexander Keller](https://research.nvidia.com/person/alex-keller), [Sanja Fidler](https://www.cs.toronto.edu/~fidler/), [高俊](https://www.cs.toronto.edu/~jungao/) <br>
* 表示共同第一作者 <br>
**[论文](https://arxiv.org/pdf/2503.03751)，[项目主页](https://research.nvidia.com/labs/toronto-ai/GEN3C/)**

摘要：本文提出GEN3C，一款兼具精准相机控制能力与时序三维一致性的生成式视频模型。现有视频生成模型虽可生成逼真视频，但往往未充分挖掘三维信息，易引发物体凭空出现或消失等全局不一致问题。即便部分模型实现了相机控制功能，其精度也十分有限——这是由于相机参数仅作为神经网络的输入，模型需自行推导视频内容与相机参数之间的依赖关系。与之不同，GEN3C以三维缓存（3D cache）为核心引导：通过预测种子图像或已生成帧的逐像素深度，得到对应的点云（point clouds）。在生成后续帧时，GEN3C以用户提供的新相机轨迹对应的三维缓存二维渲染结果作为条件输入。核心优势在于，GEN3C既无需记忆此前生成的场景内容，也无需从相机位姿中推导图像结构；借此，模型可将全部生成能力集中于未观测区域，并将场景状态推进至下一帧。实验结果表明，相较于现有工作，GEN3C可实现更精准的相机控制，且在稀疏视角新视角合成任务中达到了当前最优（state-of-the-art）性能，即便在驾驶场景、单目动态视频等挑战性场景下亦表现优异。最佳效果请通过视频查看。

## 引用

@inproceedings{ren2025gen3c,
title={GEN3C: 3D-Informed World-Consistent Video Generation with Precise Camera Control},
author={Ren, Xuanchi and Shen, Tianchang and Huang, Jiahui and Ling, Huan and
Lu, Yifan and Nimier-David, Merlin and Müller, Thomas and Keller, Alexander and
Fidler, Sanja and Gao, Jun},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
year={2025}
}