Endoscopy Artifact Detection

Detecting artifacts in endoscopy images using YOLOv2 — built during a summer internship based on the EAD2019 challenge.

What It Does

Detects 7 types of artifacts in endoscopy frames: specularity, saturation, artifact, contrast, blur, bubbles, and instruments. Uses YOLOv2 via the Darknet framework for real-time object detection.

Why YOLOv2

Compared multiple architectures (YOLO, Fast R-CNN, R-CNN). YOLOv2 had the best IoU and mAP scores, consistent with this paper.

How to Run

# Follow the Jupyter notebook
endoscopy_artifact_detection.ipynb

Download the dataset from Google Drive (2,147 annotated frames).

Tech

Python, YOLOv2, Darknet (C), OpenCV, Jupyter