Triplane Gaussian Meets LGM Convert:
Fast and Generalizable Single-View 3D Reconstruction into a mesh with Transformers

TriGSMesh combines Triplane Gaussian with an optimized version of LGM's GS to mesh conversion method to create a fast and generalizable single-view 3D reconstruction object mesh. This model is able to reconstruct 3D into a GLB format within a few minutes.

Please refer to the original repositories for more details on the reconstruction model and the conversion process:

• TriplaneGaussian
• LGM

Installation

• Python = 3.10

Please use conda to create a new environment and install the required packages:

conda create -n trigsmesh python=3.10
conda activate trigsmesh

Install pytorch using instructions from here: https://pytorch.org/get-started/locally/

Example (for CUDA 11.8):

conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia # Recommended usage

• Install pointnet2_ops

cd tgs/models/snowflake/pointnet2_ops_lib && python setup.py install && cd -

• Install pytorch_scatter

pip install git+https://github.com/rusty1s/pytorch_scatter.git

• Install nvdiffrast

pip install git+https://github.com/NVlabs/nvdiffrast

• Install nerfacc

pip install git+https://github.com/nerfstudio-project/nerfacc.git

• Install kiui's diff-gaussian-rasterization (Inria's version will not work!)

git clone --recursive https://github.com/ashawkey/diff-gaussian-rasterization
pip install ./diff-gaussian-rasterization

• Install dependencies:

pip install -r requirements.txt

• Install PyTorch3D following its official installation instruction.

Download the Pretrained Model

TriplaneGaussian offers a pretrained checkpoint available for download from Hugging Face; download the checkpoint and place it in the folder checkpoints.

from huggingface_hub import hf_hub_download
MODEL_CKPT_PATH = hf_hub_download(repo_id="VAST-AI/TriplaneGaussian", local_dir="./checkpoints", filename="model_lvis_rel.ckpt", repo_type="model")

Please note this model is only trained on Objaverse-LVIS dataset (~45K 3D models).

Inference

Use the following command to reconstruct a 3DGS model from a single image. Please update data.image_list to some specific list of image paths.

python infer.py --config config.yaml data.image_list=[path/to/image1,]
# e.g. python infer.py --config config.yaml data.image_list=[example_images/a_pikachu_with_smily_face.webp,]

The inference script will save an OBJ file with a seperated texture (not vertex colors).

Local Gradio Demo

Launch the Gradio demo locally by:

python gradio_inference.py

Acknowledgements

TriplaneGaussian

@article{zou2023triplane,
  title={Triplane Meets Gaussian Splatting: Fast and Generalizable Single-View 3D Reconstruction with Transformers},
  author={Zou, Zi-Xin and Yu, Zhipeng and Guo, Yuan-Chen and Li, Yangguang and Liang, Ding and Cao, Yan-Pei and Zhang, Song-Hai},
  journal={arXiv preprint arXiv:2312.09147},
  year={2023}
}

LGM (Convert to mesh only)

@article{tang2024lgm,
  title={LGM: Large Multi-View Gaussian Model for High-Resolution 3D Content Creation},
  author={Tang, Jiaxiang and Chen, Zhaoxi and Chen, Xiaokang and Wang, Tengfei and Zeng, Gang and Liu, Ziwei},
  journal={arXiv preprint arXiv:2402.05054},
  year={2024}
}