bag2mesh is a powerful, standalone Python tool designed to convert ROS Bag files (specifically from RGB-D cameras like Intel RealSense) into high-quality 3D Meshes (.ply).

Unlike many other tools, this does not require a ROS installation. It uses rosbags to read data directly and Open3D for advanced processing, making it lightweight and easy to run on Windows, Linux, or macOS.

• No ROS Required: Runs purely on Python.
• Robust Tracking: Implements a Local Map ICP strategy with Constant Velocity prediction to maintain tracking even during fast movements.
• Anti-Ghosting: Smart keyframe selection and strict validation logic prevent "double vision" or ghosting artifacts in the final model.
• High-Fidelity Meshing: Uses Poisson Surface Reconstruction with customizable depth and density trimming to produce watertight or clean surfaces.
• Auto-Colorization: Projects color data from the RGB camera onto the generated mesh.
• Fully Parametric: Almost every aspect of the pipeline (tracking, filtering, meshing) can be tuned via command-line arguments.

1. Clone the repository:

   git clone https://github.com/yourusername/bag2mesh.git
   cd bag2mesh

2. Install dependencies:

   pip install -r requirements.txt

   (Dependencies: open3d, rosbags, numpy, opencv-python, tqdm)

Convert a bag file to a mesh using default settings (good for most room-scale scans):

python bag2mesh.py input.bag output.ply

For scanning a small object (e.g., a chair) with high detail:

python bag2mesh.py object.bag chair.ply --voxel_size 0.01 --poisson_depth 11 --density_threshold 0.2

If scanning an object on a turntable with a fixed camera, use --max_depth to remove the background:

python bag2mesh.py turntable.bag object.ply --max_depth 0.8

To quickly check the geometry without waiting for high-res processing:

python bag2mesh.py large_scan.bag preview.ply --step 5 --voxel_size 0.05 --poisson_depth 8

For a detailed explanation of all parameters, defaults, and tuning guides, please refer to the Wiki Documentation.

• "Ghosting" (Double objects):

  • Increase --keyframe_movement_threshold (e.g., to 0.15 or 0.20).
  • Increase --voxel_size slightly.
  • Ensure --max_translation_jump is not too high.

• Mesh is too "bubbly" or fills empty spaces:

  • Increase --density_threshold (e.g., to 0.4).

• Mesh is too rough/noisy:

  • Decrease --poisson_depth (e.g., to 9 or 10).
  • Decrease --outlier_std_ratio (e.g., to 1.0).

• Tracking lost (Mesh is scrambled):

  • Increase --coarse_threshold_multiplier.
  • Increase --window_size.
  • Try processing every frame (--step 1).

MIT License