ConSeisDiff is a conditional denoising diffusion probabilistic model (DDPM) for high-fidelity synthetic seismic data generation. The framework explicitly conditions the diffusion process on geological structural priors—such as fault attributes and edge-based structural maps—to reduce the domain gap between synthetic and real seismic data while preserving reflector continuity and fault geometry.
This repository provides the official implementation accompanying our peer-reviewed publication in the Journal of Applied Geophysics, including training scripts, inference pipelines, and evaluation utilities.
Generating realistic and structurally consistent seismic data remains a major bottleneck for data-driven geophysical interpretation, particularly in fault detection and structural analysis. ConSeisDiff addresses this challenge through:
-
Structure-Aware Conditioning
Incorporation of fault attributes and edge maps to guide the diffusion trajectory toward geologically plausible solutions. -
High-Fidelity Seismic Synthesis
Generation of realistic 2D seismic sections with preserved large-scale structure and fine-scale texture. -
Quantitative Evaluation
Built-in tools for evaluating synthetic quality using metrics such as FID, SSIM, and task-oriented fault detection performance.
If you use this work, please cite the following paper:
ConSeisDiff: A Conditional Diffusion Approach to Mitigate Synthetic–Real Disparities in Seismic Fault Detection
Journal of Applied Geophysics, 243, 105956, 2025.
DOI: https://doi.org/10.1016/j.jappgeo.2025.105956
@article{FARADY2025105956,
title = {ConSeisDiff: A conditional diffusion approach to mitigate synthetic-real disparities in seismic fault detection},
journal = {Journal of Applied Geophysics},
volume = {243},
pages = {105956},
year = {2025},
issn = {0926-9851},
doi = {https://doi.org/10.1016/j.jappgeo.2025.105956},
url = {https://www.sciencedirect.com/science/article/pii/S0926985125003374},
author = {Farady, Isack and Kuo, Chia-Chen and Sellami, Soufiene and Lin, Chih-Yang}
}