Reproduce “OpenFWI: Large-Scale Multi-Structural Benchmark Datasets for Seismic Full Waveform Inversion” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Chengyuan Deng, Shihang Feng, Hanchen Wang, Xitong Zhang, Peng Jin, Yinan Feng, Qili Zeng, Yinpeng Chen and 1 more

OPENFWI introduces a comprehensive collection of 12 large-scale, multi-structural benchmark datasets totaling 2.1 TB for Full Waveform Inversion (FWI). This open-access resource facilitates the development, evaluation, and comparison of data-driven FWI algorithms, revealing current methods' limitations in generalization and complex subsurface imaging.

Key insights

  • A strong negative correlation exists between the complexity of velocity maps (measured by metrics like Shannon entropy) and the inversion accuracy of deep learning methods.
  • Current data-driven FWI methods exhibit considerable degradation in cross-domain generalization when applied to unseen datasets, highlighting a challenge for real-world applications.
  • Uncertainty in FWI predictions is significantly higher at structural boundaries and increases with higher noise levels or domain shifts, offering crucial insights into model reliability.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.