24 February 2025
14:00 Master's Defense Room 85 of IC2
With the
An Evaluation of Transformer Models for Seismic Facies Segmentation
Student
Gabriel Borges Gutierrez
Advisor / Teacher
Edson Borin
Brief summary
Since their inception, transformers have revolutionized the machine learning landscape, consistently achieving state-of-the-art performance in diverse areas such as natural language processing and image classification. However, their adoption in seismic facies segmentation has been limited, with most recent studies focusing on convolutional neural networks (CNNs). This work seeks to fill this gap by evaluating the performance of transformer-based segmentation models compared to CNN-based models. Specifically, we tested three transformers, SegFormer, Segmenter, and SetR, and two CNN models, DeepLab V3 and DeepLab V3+. The evaluation was performed using the F3 and Seam AI datasets, ensuring a fair and consistent comparison between the models. In our tests, the transformer-based model SetR consistently outperformed the CNN models, highlighting its potential for geophysical applications. Based on these promising results, we performed a hyperparameter search on SetR to further optimize its performance.
Examination Board
Headlines:
| Edson Borin | IC / UNICAMP |
| Alexandro José Baldassin | IGCE / UNESP |
| Marcelo da Silva Reis | IC / UNICAMP |
Substitutes:
| Esther Luna Colombini | IC / UNICAMP |
| Edson Takashi Matsubara | FACOM / UFMS |