seismiQB

seismiQB is a framework for research and deployment of deep learning models on post-stack seismic data. It covers all main stages of model development and its production usage for seismic interpretation. The main features are:

• Optimized IO for SEG-Y storage: by using our library segfast, combined with quantization and compression, we are working with the most popular transport format by an order of magnitude faster;
• Labelling classes, matching core interpretation entities: horizons, faults, facies, wells, and more;
• Pipelines for preparing the model inputs, e.g. patches (2d or 3d) of seismic data and corresponding segmentation masks;
• Geologic transformations, as well as more traditional ML augmentations;
• Advanced primitives for train/inference to saturate even the fastest GPUs;
• Fast and convenient export: all predicted objects are easy to convert to popular formats (SEG-Y, CHARISMA, LAS) for validation by geophysicists;

Installation

seismiQB is compatible with Python 3.8+ and well tested on Ubuntu 20.04.

# pipenv
pipenv install git+https://github.com/GeoscienceML/seismiqb.git#egg=seismiqb

# pip / pip3
pip3 install git+https://github.com/GeoscienceML/seismiqb.git

# developer version (add `--depth 1` if needed)
git clone https://github.com/GeoscienceML/seismiqb.git

Getting started

After installation just import seismiQB into your code. A quick demo of our primitives and methods:

import seismiqb

field = Field('/path/to/cube.sgy')                                    # Initialize field with SEG-Y
field.load_labels('path/to/horizons/*.char', labels_class='horizon')  # Add labeling

# Labels
field.horizons.interpolate()                                          # Fill in small holes
field.horizons.smooth_out()                                           # Smooth out and remove spikes
field.horizons.evaluate()                                             # Compute a quality control metric

# Visualizations
field.geometry.print()                                                # Display key stats about SEG-Y
field.show_slide(index=100, axis=1)                                   # Show 100-th crossline
field.show('horizons:0/metric')                                       # Show QC metric for one horizon

Be sure to check out our tutorials to get more info about the seismiQB primitives and usage.

Citing

Please cite seismiQB in your publications if it helps your research.

Khudorozhkov R., Tsimfer S., Kozhevin A., Koryagin A., Sorokina S., Strievich E. SeismiQB library for seismic interpretation with deep learning. 2023.

@misc{seismiQB_2023,
  author       = {R. Khudorozhkov and S. Tsimfer and A. Kozhevin and A. Koryagin and S. Sorokina and E. Strievich},
  title        = {SeismiQB library for seismic interpretation with deep learning},
  year         = 2023
}