VizFracSet

This repository features a dataset of fractures for computer vision algorithms, machine learning with Geological rigor.

Project tree

This dataset provides three versions for every annotation. By doing this we preserve geological context and georeferencing needed for some types of studies. For computational and statistical evaluation of ML algorithms we provide Coco Annotations and binary masks generated from the georreferenced images.

├── dataset
│   ├── vector              # QGIS proj., Georreferenced TIFFs + Shapes with fractures marked
│   ├── pixel               # PNG version of the TIFF files + Coco Annotations for fractures 
│   └── mask                # Binary png file with fractures shown as white
├── sample                  # Sample images for README
└── scripts                 # usefull python 3.x scripts to handle the dataset
    ├── convTiff_To_PNG.py  # Converts TIFF to PNG
    └── createMasksCOCO.py  # Creates mask file from PNG + Coco Annotations

Table of contents

• Requirements
• Usage
• How to cite
• Credits
• License

Requirements

To open and work with TIFF and Shapefiles:

QGIS 3.1 or above

To edit Coco Annotations you can use your favorite annotation tool, but we recommend:

Coco Annotator.

To run python scripts you should install Python 3.x and the following libs:

numpy
imageio
cv2
pycocotools.coco

To install pycocotools please run the following command:

$ pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI

Usage

This dataset allows for different usages each of the folders under dataset contains information of interest for specific groups.

In vector there are georreferenced TIFF files obtained through UAV mapping of carbonate outcrops. There is also the shapefiles that contains the lines used to delineate the fractures. All the work in this directory was curated and validated with geologists before the generation of pixel space marking or masks.

In pixel there are PNG and Coco Annotation files equivalent to the areas define in vector, files are kept with the same name in both directories, so it is possible to track which pixel space markings bellong to which original vector space marking.

In masks there are binary masks equivalent to the images in pixel

How to cite

Yet to be published.

Credits

This work is credited to the Vizlab | X-Reality and GeoInformatics Lab and the following authors and developers: Lucas Silveira Kupssinskü.

License

MIT Licence (https://mit-license.org/)