I. Introduction
As petroleum exploration continues over the decades, major oil and gas reservoir discoveries in middle or shallow basins are not expected to take place, leading to exploration toward deep basins. The prediction of the presence of the oil hosting rocks is based on the identification of different stratigraphic facies [2], which allows the distinction between distinct geological groups, essential to the comprehension of hydrocarbon migration and imprisonment. The tridimensional acquisition of seismic data results in vast amounts of data, making the manual interpretation of the seismic sections or volumes slow and biased [16], [24]. Using computational tools that support or automatize, this interpretation provides better agility and can reduce the interpreter’s subjectivism and bias. Concomitantly to modern data acquisition processes, the actual computational capacity and deep learning methods made possible the realization of many computer vision tasks using deep network architectures [12], which also has great potential in seismic interpretation.