gs_processing

A simple pipeline for point cloud processing using Open3D and DVC for reproducibility. The demo uses a point cloud scene processed from a Gaussian Splat. The pipeline has three key stages: dataloader, plane segmentation using RANSAC, and clustering using DBSCAN. All configurations are managed via the params.yaml file.

This is a dvc pipeline wrapper of the same ipynb program I leveraged for segmenting Martian lava cave point cloud using big_bot_description

Project Structure

├── src
│   ├── gs_processing
│   │   ├── pipeline
│   │   │   ├── dataloader.py  # Loads point clouds
│   │   │   ├── ransac.py      # Applies RANSAC for plane segmentation
│   │   │   ├── dbscan.py      # DBSCAN clustering
├── output                     # Stores processed output files
├── params.yaml                # Configures hyperparameters
├── dvc.yaml                   # DVC pipeline configuration
├── README.md                  # Project documentation

Requirements

• Python 3.8+
• Open3D
• DVC
• NumPy
• Matplotlib

You can install the required libraries using the following command:

pip install -r requirements.txt

Running the Pipeline

To run the pipeline, follow these steps:

1. Initialize DVC

First, initialize DVC in your project directory:

dvc init

2. Run the Pipeline

To execute the full point cloud processing pipeline, use:

dvc repro

Results

Raw Pointcloud

Clustered and Segmented pointcloud with Outlier Reduced