README.md

metadata

license: mit
size_categories:
  - 10K<n<100K

Dataset Card for Dataset Name

Homepage: https://genfusion.sibowu.com/

Repository: https://github.com/Inception3D/GenFusion?tab=readme-ov-file

Paper: GenFusion: Closing the Loop between Reconstruction and Generation via Videos

Dataset Details

Dataset that was used for training in Genfusion paper. The dataset is mostly sourced from the DL3DV-10k dataset

Dataset Description

A large-scale scene dataset, featuring 51.2 million frames from 10,510 videos captured from 65 types of point-of-interest (POI) locations, covering both bounded and unbounded scenes, with different levels of reflection, transparency, and lighting. This dataset is used to train a diffusion model to reconstruct and generate detailed 3D scenes from sparse or partial video views.

Dataset Sources [optional]

Sourced from DL3DV-10k dataset

• Repository: https://github.com/DL3DV-10K/Dataset
• Paper: DL3DV-10K: A Large-Scale Scene Dataset for Deep Learning-based 3D Vision

Uses

1. 3D scene reconstruction from monocular or multi-view video

2. Generative modeling of 3D environments

3. Sparse view synthesis and completion

Citation

Please cite the Genfusion and the DL3DV-10k paper

Genfusion

@inproceedings{Wu2025GenFusion,
    author = {Sibo Wu and Congrong Xu and Binbin Huang and Geiger Andreas and Anpei Chen},
    title = {GenFusion: Closing the Loop between Reconstruction and Generation via Videos},
    booktitle = {Conference on Computer Vision and Pattern Recognition (CVPR)},
    year = {2025}
}

DL3DV-10k

@inproceedings{ling2024dl3dv,
  title={Dl3dv-10k: A large-scale scene dataset for deep learning-based 3d vision},
  author={Ling, Lu and Sheng, Yichen and Tu, Zhi and Zhao, Wentian and Xin, Cheng and Wan, Kun and Yu, Lantao and Guo, Qianyu and Yu, Zixun and Lu, Yawen and others},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={22160--22169},
  year={2024}
}