HoloPart: Generative 3D Part Amodal Segmentation

Generative 3D part amodal segmentation--decomposing a 3D shape into complete, semantically meaningful parts.

🔥 Updates

📅 April 2025

• 🚀 Initial Release: Published code, pretrained models, and interactive demo.
• 📌 Coming Soon:
• [ ] Integration of segmentation methods into the HoloPart pipeline.

🔨 Installation

Clone the repo:

git clone https://github.com/VAST-AI-Research/HoloPart.git
cd HoloPart

Create a conda environment (optional):

conda create -n holopart python=3.10
conda activate holopart

Install dependencies:

# pytorch (select correct CUDA version)
pip install torch torchvision --index-url https://download.pytorch.org/whl/{your-cuda-version}

# other dependencies
pip install -r requirements.txt

💡 Quick Start

Step 1: Prepare segmented mesh

Upload a mesh with part segmentation. We recommend using these segmentation tools: - SAMPart3D - SAMesh

For a mesh file mesh.glb and corresponding face mask mask.npy, prepare your input using this Python code:

import trimesh
import numpy as np
mesh = trimesh.load("mesh.glb", force="mesh")
mask_npy = np.load("mask.npy")
mesh_parts = []
for part_id in np.unique(mask_npy):
    mesh_part = mesh.submesh([mask_npy == part_id], append=True)
    mesh_parts.append(mesh_part)
mesh_parts = trimesh.Scene(mesh_parts).export("input_mesh.glb")

The resulting input_mesh.glb is the prepared input for HoloPart.

Step 2: Decompose the 3D mesh into complete parts:

python -m scripts.inference_holopart --mesh-input assets/example_data/000.glb

The required model weights will be automatically downloaded: - HoloPart model from VAST-AI/HoloPart → pretrained_weights/HoloPart

⭐ Acknowledgements

We would like to thank the following open-source projects and research works that made HoloPart possible:

• 🤗 Diffusers for their excellent diffusion model framework
• HunyuanDiT for DiT
• FlashVDM for their lightning vecset decoder
• 3DShape2VecSet for 3D shape representation
• TripoSG as our base model

We are grateful to the broader research community for their open exploration and contributions to the field of 3D generation.

📚 Citation

@article{yang2025holopart,
      title={HoloPart: Generative 3D Part Amodal Segmentation}, 
      author={Yang, Yunhan and Guo, Yuan-Chen and Huang, Yukun and Zou, Zi-Xin and Yu, Zhipeng and Li, Yangguang and Cao, Yan-Pei and Liu, Xihui},
      journal={arXiv preprint arXiv:2504.07943},
      year={2025}
}