Synthetic Data Generation Pipeline

A Synthetic Data Generation Pipeline (SDGP) that can generate synthetic training data for various neural networks (currently: ROCA) using CAD models in STEP format and NVIDIA Isaac Sim.

Prerequisites

• NVIDIA RTX GPU
• Installed NVIDIA driver
• Anaconda / Miniconda

Install Miniconda

• Download the Anaconda installer

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh

• Run the installer

bash ~/Miniconda3-latest-Linux-x86_64.sh -b -u

• Finish the installation and reload the terminal

source ~/.bashrc

Set up the Conda environment

conda env create -f environment.yaml

Using the SDGP

• Place CAD models in STEP format into the directory ~/custom_dataset/cad_models/step
• Run the pipeline

./create_dataset.sh

• The dataset for ROCA will be located at ~/custom_dataset/converted/6_dof/ReplicatorToROCA
• The working directory is defined in the Bash script via the variable CUSTOM_DATASET_DIR, which points to ~/custom_dataset. If a different path should be used, this variable can be modified.

Development

conda env create -f environment.yaml
conda activate sodah-sdgp
python -m isaacsim --generate-vscode-settings

Sources

• https://docs.isaacsim.omniverse.nvidia.com/5.0.0/installation/install_python.html (12 Aug 2025)

Reproducing the Results of "Synthetic Data Generation Pipeline for CAD-Based Object Reconstruction and Pose Estimation"

Instructions can be found in the directory paper-reproduction.