Skip to content

Trying to use transfer learning and GANs to build a system that will recognize polyps in colonoscopy footage with accuracy of >98%

Notifications You must be signed in to change notification settings

McClain-Thiel/polyp_vis

Repository files navigation

Polyp Vision

This is a project I worked on in the summer between my sophmore and junior year. It was a project that was dropped from a company I was interning at and I picked it up. The original idea was a tool for medical professionals performing colonoscopies that would localize and tag polyps in real time. This repo is just me experimenting with real time image processing and was my first real exprience with computer vision so its both poorly engineered and a bit simplistic. While not fully optimized or anything, I did manage to use some combonation of transfer learning, YOLOv3, and possibly some RCNN to build a system that worked, if very slowly. Below is a short gif of the system tagging polyps in single frames. I didn't have the processing power or expirience to build an actual live model that would work on videos.

Demo

About

Trying to use transfer learning and GANs to build a system that will recognize polyps in colonoscopy footage with accuracy of >98%

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published