Automated Chondrite Segmentation Pipeline

This repository contains the code, notebooks, and resources for the automated segmentation and analysis of chondritic meteorites, as described in our report. The pipeline combines BSE imaging and EDS elemental maps with a patch-based convolutional neural network (PatchCNN) and an object-wise segmentation refinement.

Repository Structure

.
├── README.md
├── notebook           # Jupyter notebooks for experimentation and analysis
├── resources          # Weights, project files, shaders, and styles
├── viewer             # Interactive user interface
├── pyproject.toml     # Project configuration
├── run.sh             # Script to launch the pipeline
└── uv.lock

Interactive User Interface

The features are:

• Visualize multiple synchronized views of the same meteorite region, each showing complementary information:

  • BSE map (structural reference)
  • EDS-derived elemental maps (MgSi, MgCaAl, FeSO)
  • Label map (manual annotations)
  • PatchCNN predictions
  • Object-wise segmentation map

• Annotate: Each class has a unique color, brush size is configurable, and annotations are synchronized across all views.

• Save labels for further analysis or retraining.

• Inference of the PatchCNN and the object-wise pipeline.

Below is an example screenshot of the viewer:

Resources

• Pre-trained model weights: resources/lcnn_weights.pth
• Hyperparameters of the model: resources/hp.json
• Project configuration: resources/project*.json

Project Configuration (project*.json)

The project*.json file defines the images, views, and features used by the pipeline. Example:

{
  "imgs": { "BSE": "Mosaic_BSE.tif", "Mg": "Mosaic_Mg.tif", "Si": "Mosaic_Si.tif" },
  "views": {
    "cols": 2, "rows": 2,
    "datas": [
      { "col": 0, "row": 0, "grayscale": true,  "r": "BSE" },
      { "col": 1, "row": 0, "grayscale": false, "r": "Mg", "g": "Si", "b": "0" }
    ]
  }
}

• imgs: maps channel names to image files.
• views: defines the viewer grid and which channels appear in each view (red/green/blue, or 0 if empty).

Running the Pipeline

# Install dependencies
uv sync

# Launch the UI
./run.sh