3D Seismic–Well Multimodal Representation

This project explores how to integrate dense 3D seismic fields with sparse well-log observations in a shared spatial representation.

Current direction

The current pipeline investigates Gaussian feature fields, local seismic tokens, well-point features, structure-aware cross-modal attention and spatial priors derived from neighboring wells.

Research questions

• How should dense volumetric observations interact with sparse point-wise measurements?
• Which information should be aligned globally, and which should be exchanged locally?
• Can a continuous 3D representation make cross-modal querying more natural?