RobustVGGT

<p align="center"> <h1 align="center">Emergent Outlier View Rejection in Visual Geometry Grounded Transformers</h1> <p align="center"> <a href="https://onground-korea.github.io/">Jisang Han</a>^1,2* · <a href="https://sunghwanhong.github.io/">Sunghwan Hong</a>^3* · <a href="https://crepejung00.github.io/">Jaewoo Jung</a>¹ · <a href="https://scholar.google.com/citations?hl=ko&user=7cyLEQ0AAAAJ">Wooseok Jang</a>¹ · <a href="https://hg010303.github.io/">Honggyu An</a>¹ · <a href="https://qianqianwang68.github.io/">Qianqian Wang</a>⁴ · <a href="https://scholar.google.com/citations?user=cIK1hS8AAAAJ">Seungryong Kim</a>^1† · <a href="https://scholar.google.com/citations?user=YeG8ZM0AAAAJ">Chen Feng</a>^2† </p> <h4 align="center">¹KAIST AI, ²New York University, ³ETH AI Center, ETH Zurich, ⁴UC Berkeley</h4> <!-- <p align="center">^‡Work done during a visiting researcher at New York University&emsp;^*Equal contributions&emsp;^†Co-corresponding</p> --> <h3 align="center"> <a href="https://arxiv.org/abs/2512.04012">arXiv</a> | <a href="https://github.com/cvlab-kaist/RobustVGGT/releases/download/paper/Emergent.Outlier.View.Rejection.in.Visual.Geometry.Grounded.Transformers.pdf">Paper (High quality)</a> | <a href="https://cvlab-kaist.github.io/RobustVGGT">Project Page</a> </h3> <h3 align="center"> CVPR 2026 </h3> </p> <p align="center"> <a href=""> <img src="assets/teaser.png" alt="Logo" width="80%"> </a> </p>

We reveal that Visual Geometry Grounded Transformers (VGGT) has a built-in ability to detect outliers, which we leverage to perform outlier-view rejection without any fine-tuning.

What to expect:

• [x] Demo inference code
• [ ] Evaluation code
• [ ] Visualization code

Installation

Our code is developed based on pytorch 2.5.1, CUDA 12.1 and python 3.10.

We recommend using conda for installation:

conda create -n robust_vggt python=3.10
conda activate robust_vggt

pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

Running Demo

To run the robust reconstruction demo with outlier rejection:

python robust_vggt.py --image-dir examples/trevi

python robust_vggt.py --image-dir examples/notredame --rej-thresh 0.3

Citation

@article{han2025emergent,
  title={Emergent Outlier View Rejection in Visual Geometry Grounded Transformers},
  author={Han, Jisang and Hong, Sunghwan and Jung, Jaewoo and Jang, Wooseok and An, Honggyu and Wang, Qianqian and Kim, Seungryong and Feng, Chen},
  journal={arXiv preprint arXiv:2512.04012},
  year={2025}
}

Acknowledgement

We thank the authors of VGGT for their excellent work and code, which served as the foundation for this project.