Models - U-Net and SegFormer for Automated Fracture Detection trained on FraXet
Model Details
Model types:
U-Net: convolutional encoder–decoder with skip connections [Ronneberger et al. 2015]
SegFormer: transformer-based encoder with lightweight MLP decoder [Xie et al. 2021 ] (implemented by smp)
License: MIT
Dataset: FraXet (zenodo)
Repository: github.com/ayoubft/fractex2d.pt
Demo: huggingface.co/spaces/ayoubft/fractex2d
Paper: coming soon
Model Description
These models perform pixel-wise fracture segmentation from paired RGB and DEM patches of outcrop imagery.
They serve as baseline architectures in the FraXet benchmarking framework comparing classical filters, CNNs, and transformer models for geological fracture mapping.
Uses
Direct use: Predict fracture probability maps or binary masks for UAV or field imagery (RGB + DEM).
Downstream use: Use as baseline models or assistive pre-annotation tools for geoscience datasets.
Bias, Risks, and Limitations
Predictions depend on annotation quality, illumination, and lithology.
Thin or poorly illuminated fractures may be missed; shadows and texture can yield false positives.
Use predictions as assistive probability maps and validate with expert interpretation.