---
license: mit
library_name: pytorch
tags:
- computer-vision
- 3d-gaussian-splatting
- gaussian-splatting
- super-resolution
- eccv-2026
- pytorch
pipeline_tag: image-to-3d
---

# AnchorSplat 20x Checkpoint

This repository hosts the released 20x checkpoint for **AnchorSplat: Fast and Structure Consistent Detail Synthesis for Gaussian Splatting (ECCV 2026)**.

- Paper: https://arxiv.org/abs/2607.01290
- Code: https://github.com/zhude233/AnchorSplat
- Checkpoint: `anchorsplat_20x.pth`
- SHA256: `ad05f8b965c002c1f62cea53e4ce10ed4804bbc433375afa5f411f236d1b79a3`

## Usage

Download the checkpoint and place it at:

```text
checkpoints/anchorsplat_20x.pth
```

or pass it explicitly:

```bash
WEIGHTS=/path/to/anchorsplat_20x.pth \
  bash scripts/inference_external.sh examples/lgm_sample.ply outputs/lgm_sample_refined.ply lgm
```

## Notes

AnchorSplat is designed for fast, generalizable, plug-and-play enhancement of low-quality 3D Gaussian Splatting assets. For external PLY inputs, please follow the input format and normalization instructions in the GitHub README.

## Citation

```bibtex
@article{zhu2026anchorsplat,
  title={AnchorSplat: Fast and Structure Consistent Detail Synthesis for Gaussian Splatting},
  author={Zhu, Dexu and Shao, Jiangnan and Wang, Xiaofeng and Duan, Junxian and Cao, Jie and Zhu, Zheng and Huang, Huaibo},
  journal={arXiv preprint arXiv:2607.01290},
  year={2026},
  eprint={2607.01290},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  doi={10.48550/arXiv.2607.01290},
  url={https://arxiv.org/abs/2607.01290}
}
```