HorizonRobotics/3D-Fixer-eval-data

---
license: apache-2.0
task_categories:
- image-to-3d
---

# ARSG-110K

ARSG-110K is a large-scale scene-level dataset for 3D scene generation and completion, introduced in the paper [3D-Fixer: Coarse-to-Fine In-place Completion for 3D Scenes from a Single Image](https://huggingface.co/papers/2604.04406).

- **Project Page:** [https://zx-yin.github.io/3dfixer](https://zx-yin.github.io/3dfixer)
- **Github Repository:** [https://github.com/HorizonRobotics/3D-Fixer](https://github.com/HorizonRobotics/3D-Fixer)
- **Paper:** [https://arxiv.org/abs/2604.04406](https://arxiv.org/abs/2604.04406)

## Dataset Summary

ARSG-110K is designed to address the data scarcity bottleneck in compositional 3D scene generation. It is one of the largest scene-level datasets to date, comprising:
- Over **110,000** diverse scenes.
- **3 million** annotated images.
- High-fidelity 3D ground truth, including object-level layouts and annotations.
- Built upon the [TRELLIS-500K](https://github.com/microsoft/TRELLIS/blob/main/DATASET.md) dataset.

The dataset supports tasks such as in-place completion, where complete 3D assets are generated conditioned on partially visible point clouds cropped from fragmented geometry.

## Citation

```bibtex
@inproceedings{yin2026tdfixer,
title={3D-Fixer: Coarse-to-Fine In-place Completion for 3D Scenes from a Single Image},
author={Yin, Ze-Xin and Liu, Liu and Wang, Xinjie and Sui, Wei and Su, Zhizhong and Yang, Jian and Xie, jin},
booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
year={2026}
}
```