Project 2: GANs for Automatic Polyp Segmentation

Task

Polyps are one of the causes of colorectal cancer and early diagnosis of polyps by can lead to successful treatment. The details of the task is in Medico Challenge website.


Action: Developed GANs based Model

My contribution shows a use of GANs-based model to automatically segmenting out the polyps area within gastrointestinal images. Perceiving the problem as an image-to-image translation task, conditional generative adversarial networks are utilized to generate masks conditioned by the images as inputs. Both generator and discriminator are convolution neural networks based.


Result: 

I was not in the leaderboard, however, a working paper is published, showing more details. Mainly the work contributed in bench-marking the automatic polyp segmentation. The model achieved 0.4382 on Jaccard index and 0.611 as F2 score.

Some Good Examples  😊

Some Bad Examples  😏


More details about: Generative Adversarial Networks for Automatic Polyp Segmentation

Publication: 


A. Ahmed, Awadelrahman M. "Generative Adversarial Networks for Automatic Polyp Segmentation" Proceedings of the MediaEval 2020 Workshop https://www.eigen.no/ , 14-15 December 2020.


Some experimental code on my GitHub:

https://github.com/Awadelrahman/GANPolypSeg

Video presentation: 

Server IP: 50.18.238.17