EAGE-Annual-Hackathon / SnorIAX

EAGE Hackathon 2022

SnorIAX | EAGE Hackathon 2022

Explainable AI repository for Porous media Generation (Discrimiantor Evaluation) and Salt Segmentation.

Main Objectives

• Use saliency and oclussion approaches to explain what the discrimanator is doing in a GAN example for porous media generation.
• Use attention maps alongside the training for a salt segmentation problem.

Table of Contents

Main Folders

This repository contains the following folders:

📂 Porous media:

• It includes the models for each epoch (Generator and Discrimator) for a WGAN with gradient penalty for the Berea and Beadpack dataset. Also, it provides the visualization using saliency and oclussion approaches.
• Data needed for PorousMedia GAN: https://drive.google.com/drive/folders/1Td-pGOgEzdyNLc9a53ZH20tzxfq2GwRq?usp=sharing

📂 AttentionSaltNet

• Salt segmentation problem focused on attention maps.

📂 saltNet

• Not Successfull approach to apply saliency and oclussion approaches :( sadly we did not manage to obtain good results.

Slides

The slides can be found in pptx. format or in drive in the following link https://docs.google.com/presentation/d/1IcY3HfhFEcwlcRbES3Sbveordv7SDEVzcrBM5VCnJpI/edit?usp=sharing