Download Data Codes Participants Call
CRACKS is a crowdsourced seismic fault dataset curated through Amazon Mechanical Turk (MTurk), spanning multiple levels of expertise and confidence.
The Annotations and Folder Structure
Our data is constructed by crowdsourcing 35 annotators across 3 levels of expertise:
- 1 expert: a domain specialist who has both academic training in geophysics and has been working in the field for several years
- 8 practitioners: people who have some degree of familiarity with this data, but don’t possess any formal background in seismic interpretation
- 26 novices: people with no prior exposure to seismic data, guided only by our 5-minute instructional video
Each annotator also makes use of 3 available labels to detect their faults:
- No fault (certain): Annotator is sure there are no faults in this region. Color coded orange, value 1
- Fault (uncertain): Annotator believes there might be a fault, but is not entirely sure. Color coded green, value 2
- Fault (certain): Annotator is sure there is a fault. Color coded blue, value 3
All the fault masks are saved as png files, following the naming convention of the corresponding image (i.e. section_001.png, up to section_400.png).
We have organized our folder structure in terms of expertise levels, i.e. {exp_level}{annotator_id}/section_{3-digit slice index}.png, for example novice19/section_008.png.
Ethics and Data Collection Process
The data collection process was deemed to be Minimal risk research qualified for IRB exemption status under 45 CFR 46 104d.2 (Protocol H23360). All annotators are between 18 – 89 years old, and are able to consent without requiring surrogate consent from a legally authorized representative.
We leverage the MTurk platform for our labeling task, adapting existing image segmentation layouts to suit our needs. We establish 2 bonuses to ensure retention and focus, both by rewarding completion of the annotation process with a prorated pay bonus, and by rewarding self-consistency in redundant quality-assesment (QA) images.
Citation and Usage
BibTeX
@data{11559387
doi = {10.5281/zenodo.11559387},
url = {https://zenodo.org/records/11559387},
author = {Mohit Prabhushankar, Kiran Kokilepersaud, Jorge Quesada, Yavuz Yarici, Chen Zhou, Mohammad Alotaibi, Ghassan AlRegib, Ahmad Mustafa, Yusufjon Kumakov},
publisher = {Zenodo},
title = {CRACKS: Crowdsourcing Resources for Analysis and Categorization of Key Subsurface faults}, year = {2024} }
