RockCuttingsCharacterization

The goal of this project is to guess the rock type from CT-scan cutting with deep learning.

It is based on @nfholsen's repository Rock_Cuttings_Characterisation with emphasis on making it easy to use, up to date and running. The PIXE lab at EPFL has the pre-processed images.

The model is trained on the visible minerals (after having preprocessed the CTscans then the rock cuttings) that look like the following:

The project layout is as follows:

• config: where everything relative to the model parameters and dataset combinations is
  • sets: *.csv files containing the paths to the chosen images in the Data folder
• data: where all images are kept, grouped by origin
  • Borehole: in-situ samples
  • Lab: In laboratory scanned cuttings
  • Lab_New: contains all of Lab images and some new ones
  • New: where samples are saved when pre-processing
  • Raw: where the raw (not preprocessed yet) images are kept
  • Train: A combination of the Borehole and Lab samples since not every type is available from the Lab set
• outputs: model results will be saved here, sorted by model name defined in the config (*.yaml)) files
• report: image examples and plots
• scripts
  • Utils: modules to lighten the main scripts. Note: the path_to function is particularily useful as it takes a relative path with the project as root and transforms it into an absolute path. All paths in the codes should be relative or used via path_to()
  • *.py executables and modules
  • main.ipynb for any purposes
  • scripts.log to help understand and debug the project

Step-by-step usage: suppose we have a new set of untreated new images

1. Pre-process the files using the preprocess.py script. You can either call its function preprocess(path_to_raw_folders, folder_to_save_in) or run the command: python3 scripts/preprocess.py -i data/Raw -o data/New

2. Write a *.csv file containing the image paths and labels or multiple files if you want to split them into train and validations sets already. The script make_sets.py I used could be useful. It should be saved in config/sets

3. Write the *.yaml config file, you can use the config/debugconfig.yaml file as a template

4. Train a model by calling the train() function in scripts/train.py or run the command:

   python3 scripts/train.py config/config_file.yaml

   The outputs with results logs should appear under outputs/model_name

5. Test your model with scripts/test.py