Fractex2D – Fracture Detection

Try out deep models that use RGB+DEM inputs along with classic vision methods that work on RGB images. Support for RGB-only deep models is on the way.

Deep Learning Segmentation

Patch-based fracture segmentation using UNet or SegFormer trained on FraXet dataset.

Requirements before running:

  • RGB image: .png or .tif
  • DEM: .tif (GeoTIFF — all spatial metadata is preserved in the output)
  • Both must have same resolution

The model processes the RGB + DEM pair in 256×256 patches internally to produce a binary fracture map, while still allowing you to input images of any size.
The output GeoTIFF inherits every metadata tag from the DEM (CRS, transform, tiling, compression, …) with only dtype and band count adapted.
Model
Examples
RGB image (.png/.tif) DEM (.tif) Model

The sample images included with this interface originate from: Nordbäck, N., & Ovaskainen, N. (2022). UAV-acquired orthomosaics of Loviisa shoreline outcrops (Version 1.0.0) [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.7077519