[ICCV 2025] Official implementation of PRM, a feed-forward framework for high-quality 3D mesh generation with photometric stereo images.

• [✅] Release inference and training code.
• [✅] Release model weights.
• [✅] Release huggingface gradio demo. Please try it at demo link.
• Release ComfyUI demo.

We recommend using Python>=3.10, PyTorch>=2.1.0, and CUDA>=12.1.

conda create --name PRM python=3.10
conda activate PRM
pip install -U pip

# Ensure Ninja is installed
conda install Ninja

# Install the correct version of CUDA
conda install cuda -c nvidia/label/cuda-12.1.0

# Install PyTorch and xformers
# You may need to install another xformers version if you use a different PyTorch version
pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu121
pip install xformers==0.0.22.post7

# Install Triton 
pip install triton

# Install other requirements
pip install -r requirements.txt

The pretrained model can be found model card.

Our inference script will download the models automatically. Alternatively, you can manually download the models and put them under the ckpts/ directory.

bash run.sh

We provide our training code to facilitate future research. For training data, we used filtered Objaverse for training. Before training, you need to pre-processe the environment maps and OBJ files into formats that fit our dataloader.

# OBJ files to mesh files that can be readed
python obj2mesh.py path_to_obj save_path

For preprocessing environment maps, please run

# Pre-process environment maps
python light2map.py path_to_env save_path

To train the sparse-view reconstruction models, please run:

# Training on Mesh representation
python train.py --base configs/PRM.yaml --gpus 0,1,2,3,4,5,6,7 --num_nodes 1

Note that you need to change to root_dir and light_dir to pathes that you save the preprocessed GLB files and environment maps.

If you find our work useful for your research or applications, please cite using this BibTeX:

@article{ge2024prm,
  title={PRM: Photometric Stereo based Large Reconstruction Model},
  author={Ge, Wenhang and Lin, Jiantao and Shen, Guibao and Feng, Jiawei and Hu, Tao and Xu, Xinli and Chen, Ying-Cong},
  journal={arXiv preprint arXiv:2412.07371},
  year={2024}
}

We thank the authors of the following projects for their excellent contributions to 3D generative AI!

• FlexiCubes
• InstantMesh