2.5D Transformer: An Efficient 3D Seismic Interpolation Method without Full 3D Training

TL;DR

This work tackles the computational bottleneck of applying Transformer-based interpolation to 3D seismic data by introducing a 2.5D Transformer (T-2.5D) that leverages cross-dimensional transfer learning. A two-stage training pipeline—2D pre-training of 2D Transformer encoders followed by 3D fine-tuning of 3D Seismic Dimension Adapters (SDAs)—enables 3D-aware interpolation without full 3D training. Empirical results on Kerry, Parihaka, and Opunake datasets show that T-2.5D achieves interpolation performance comparable to a full 3D Transformer while reducing memory usage and training time significantly. The approach offers a practical, scalable path to high-quality 3D seismic interpolation with Transformer architectures, suitable for large-scale geophysical datasets.

Abstract

Transformer has emerged as a powerful deep-learning technique for two-dimensional (2D) seismic data interpolation, owing to its global modeling ability. However, its core operation introduces heavy computational burden due to the quadratic complexity, hindering its further application to higher-dimensional data. To achieve Transformer-based three-dimensional (3D) seismic interpolation, we propose a 2.5-dimensional Transformer network (T-2.5D) that adopts a cross-dimensional transfer learning (TL) strategy, so as to adapt the 2D Transformer encoders to 3D seismic data. The proposed T-2.5D is mainly composed of 2D Transformer encoders and 3D seismic dimension adapters (SDAs). Each 3D SDA is placed before a Transformer encoder to learn spatial correlation information across seismic lines. The proposed cross-dimensional TL strategy comprises two stages: 2D pre-training and 3D fine-tuning. In the first stage, we optimize the 2D Transformer encoders using a large amount of 2D data patches. In the second stage, we freeze the 2D Transformer encoders and fine-tune the 3D SDAs using limited 3D data volumes. Extensive experiments on multiple datasets are conducted to assess the effectiveness and efficiency of T-2.5D. Experimental results demonstrate that the proposed method achieves comparable performance to that of full 3D Transformer at a significantly low cost.

Figures (17)

• Figure 1: Architecture of Transformer. (a) Transformer encoder for 2D features, and (b) calculation process of MSA.
• Figure 2: Illustration of the T-2D. (a-c) Architectures of T-2D, HB and TB, respectively.
• Figure 3: Architecture of the T-3D, showing the changes in the dimensions of feature maps from 40×40 to 40×40×16 and network components from 2D to 3D.
• Figure 4: (a) and (b) Architectures of the T-2.5D and SDA, respectively.
• Figure 5: Cross-dimensional TL of the T-2.5D.