Readme.md

August 14, 2025 · View on GitHub

One View, Many Worlds: Single-Image to 3D Object Meets Generative Domain Randomization for One-Shot 6D Pose Estimation

arXiv Project Page

Zheng Geng, Nan Wang, Shaocong Xu, Bohan Li, Zhaoxi Chen, Chongjie Ye, Sida Peng, Hao Zhao


Manipulation


Instruction


TODO:

  • Release Polished Paper
  • Release Evaluation Code
  • Release Training Code

⚙️ Installation

Quick Setup Using setup.sh

To streamline the setup process, we provide a setup.sh script that automates all installation steps. Follow these instructions to get started:

Prerequisites

Before running the setup.sh script, ensure you have the following prerequisites installed on your system:

  • Python 3.11+
  • Conda (recommended)
  • Basic build tools (git, make, cmake, etc.)

Usage

  1. Clone the repository:

    git clone https://github.com/GZWSAMA/OnePoseviaGen.git
    cd OnePoseviaGen
    
  2. Make the setup.sh script executable:

    chmod +x setup.sh
    
  3. Run the setup.sh script:

    It is recommended to run this script within a fresh conda environment. Here's how you can create and activate a new environment before running the script:

    conda create -n OnePoseviaGen python=3.11 -y
    conda activate OnePoseviaGen
    ./setup.sh
    

    The script will handle:

    • Installing PyTorch with CUDA support.
    • Installing required dependencies.
    • Cloning and installing external extensions.
    • Building FoundationPose.
    • Installing local packages in editable mode.
    • Downloading pretrained weights.
    • Patching the transformers library.
  4. Verify the setup:

    After running the script, verify that all dependencies are correctly installed and the necessary files are downloaded.

  5. Web demo:

    python app.py
    

Troubleshooting

Indices should be on the same device

Can't convert cuda:0 device type tensor to numpy


📚 Citation

If you find this work useful, please cite:

TODO