Author: Harsh Kumar
Date of completion: 03/08/2019
It was my Summer internship work to detect different type of artifact in endoscopy images, it was based on EAD2019 challenge.
-
Read several research paper and analyse different type of object detection algorithms eg. YOLO, fastRcnn, Rcnn etc.
-
Implement and analyse the result and compare with the research papers.
-
Analyse the matrix such and iou,mAP and loss.
To complete my task i have used YOLOv2 because it's iou,mAP was more as suggested in the research paper. link: https://arxiv.org/pdf/1904.07073.pdf To implement object detection model i have darknet framwork for object detection which is written in c.
Dataset was provided by my mentors which can be downoad from my google drive link: https://drive.google.com/open?id=1IFliSsmF_Srr0M5psR_XgF0uUK0Hdr5Y
About dataset: The training dataset for detection consists in total 2147 annotated framesover all 7 artifact classes (Specularity, Saturation, artifact, contrast, blur, bubbles, instrument) more info about dataset: https://arxiv.org/pdf/1905.03209.pdf
Follow jupyter notebook: endoscopy_artifact_detection.ipynb.
"slice_script.py" : extract usefull data from output2 folder such as(iou,map etc).
"process.py": generate train txt file and test txt file.
"easy_install_1.py", "easy_install_2.py": script to install required dependencies.