| Texto completo | |
| Autor(es): |
Marques, Fernando G.
;
Astudillo, Carlos A.
;
Souza, Alan
;
Miranda, Daniel
;
Borin, Edson
Número total de Autores: 5
|
| Tipo de documento: | Artigo Científico |
| Fonte: | IEEE Geoscience and Remote Sensing Letters; v. 23, p. 5-pg., 2026-01-01. |
| Resumo | |
We apply the self-supervised learning (SSL) technique of vision transformers masked autoencoder (ViTs MAE) models with the goal of producing a feature extractor ViT backbone for neural networks that receive seismic data as an input. We then evaluate the quality of these backbones by coupling them to a simple linear prediction head and fine-tuning these models in a seismic semantic segmentation task. We compare domain-specific ViT MAE against cross-domain pretrained and randomly initialized ViTs and show that it yields superior performance in low-data regimes. We also demonstrate that pretraining loss correlates with downstream performance, supporting its use as a proxy for feature quality. (AU) | |
| Processo FAPESP: | 13/08293-7 - CECC - Centro de Engenharia e Ciências Computacionais |
| Beneficiário: | Munir Salomao Skaf |
| Modalidade de apoio: | Auxílio à Pesquisa - Centros de Pesquisa, Inovação e Difusão - CEPIDs |