UFNO Seismic Reconstruction
Learning a robust 3D reconstruction operator
About
UFNO Seismic Reconstruction contains the competition and conference workflow for learning a 3D reconstruction operator with U-FNO (Traversa et al. 2025). It trains on patches from open 3D field seismic cubes and considers random, trace-wise, and regular missing-data patterns.
Why it matters
Acquisition gaps and failed receivers leave missing seismic traces that degrade later processing and interpretation. Operator learning seeks a reusable reconstruction map that generalizes across missing ratios and data resolutions.
Key result and audience
The project provides data-preparation, training, pretrained-weight, and notebook guidance developed for the CPG-KFUPM machine-learning competition and an EAGE conference paper. It is useful to researchers benchmarking 3D seismic reconstruction and neural operators.
Public links
This public SEEM-KFUPM repository is a fork of the competition project and is licensed under GPL-3.0.