seisbench/seisbench

What it solves

SeisBench reduces the overhead associated with developing and applying machine learning techniques to seismology. It provides a standardized way to access seismic datasets and a unified API for training and deploying ML models on seismic data, eliminating the need to build custom data pipelines for every new project.

How it works

The toolbox is organized into three core modules:

• data: Provides a unified interface for loading and accessing benchmark seismic datasets.
• models: Contains a collection of machine learning models for seismology, allowing users to create new models, load pretrained ones, or train existing models on specific datasets.
• generate: Bridges the gap between data and models by providing tools to build data generation pipelines.

Who it’s for

Seismologists and machine learning researchers who need to process seismic data, train models for tasks like earthquake picking, denoising, or depth estimation, and work with distributed acoustic sensing (DAS) data.

Highlights

• Unified API: Simplifies the process of accessing datasets and applying ML algorithms.
• Pretrained Models: Offers a collection of ready-to-use models for seismological tasks.
• DAS Support: Includes specific tools and models for distributed acoustic sensing data.
• Comprehensive Examples: Provides Colab notebooks for various tasks, including event catalog building and denoising.