You are on page 1of 6

TGS Salt

22 Feb 2019
Competition

• Several areas of Earth with large accumulations of oil and gas also have huge
deposits of salt below the surface.
• Professional seismic imaging still requires expert human interpretation of salt
bodies. This leads to very subjective, highly variable renderings.
• The goal is to build and algorithm that automatically and accurately identifies if a
subsurface target is salt or not on seismic images.
Dataset

The dataset contains 101x101 images with each pixel classified as salt
or not. Images are seismic reflection images taken from some locations,
and mask images are showing where the salt is (the white part).
Scoring by Intersection over Union (IoU)
IoU = Area of Intersection / Area of Union
The “Area of Intersection” is the area where our predicted image overlaps the ground truth image
(the actual “salt mask”).
The “Area of Union” is the area combining the predicted image and the ground truth image.
Here IoU is 872 / 1349 = 0.65. A more accurate detection will have a value toward 1.
U-Net architecture
Further reading
• U-Net original article
https://arxiv.org/abs/1505.04597

• 14th place solution


https://nikolasent.github.io/deeplearning/competitions/2018/10/24/S
emantic-Segmentation-of-Seismic-Reflection-Images.html

You might also like