An open API service indexing awesome lists of open source software.

https://github.com/adamdad/threestudio-lrm


https://github.com/adamdad/threestudio-lrm

Last synced: about 1 year ago
JSON representation

Awesome Lists containing this project

README

          

# threestudio-lrm
The threestudio-lrm is an extension for threestudio, integrating the Large Reconstruction Model (LRM) for advanced 3D reconstruction tasks. This model is based on the framework detailed at [https://yiconghong.me/LRM/](https://yiconghong.me/LRM/). We have adapted the open-source implementation from [OpenLRM](https://github.com/3DTopia/OpenLRM). Currently, it is only used for initialization of Gaussian Splatting.

## Installation
```
cd custom
git clone https://github.com/Adamdad/threestudio-lrm

# install openLRM
cd threestudio-lrm
git clone https://github.com/3DTopia/OpenLRM.git
cd OpenLRM
pip install -r requirements.txt
```

## Examples
Please see [threestudio-3dgs](https://github.com/DSaurus/threestudio-3dgs#load-from-ply) for more details.

## Supported Modes

The threestudio-lrm currently supports two modes:

1. **Text-to-Image Mode (`text2image`):**
- Generates a single image from a text prompt using SDXL.
- Transforms the image into a triplanar representation, then to a mesh, and finally to a point cloud using LRM.
- Initialize the 3DGS

2. **Image-to-Image Mode (`image2image`):**
- Directly loads a single-view image.
- Generates a point cloud from this image.
- Initialize the 3DGS

## Citation
If you use threestudio-lrm in your research, please cite the following paper:
```
@article{hong2023lrm,
title={Lrm: Large reconstruction model for single image to 3d},
author={Hong, Yicong and Zhang, Kai and Gu, Jiuxiang and Bi, Sai and Zhou, Yang and Liu, Difan and Liu, Feng and Sunkavalli, Kalyan and Bui, Trung and Tan, Hao},
journal={arXiv preprint arXiv:2311.04400},
year={2023}
}
```