PolarNet, GC (Glandular Cell) detection from cytology WSI

Code available of a preprint paper Cervical Glandular Cell Detection from Whole Slide Image with Out-Of-Distribution Data (refer to arXiv). Please cite it.

Plugin PolarNet using eight neighboring self-attention mechanism, for improving GC (glandular cell) detection from cytology WSI (Whole Slide Image).

It can be plugged into any modern deep learning-based object detection model. The code of PolarNet polar_net.py in this project is easiest to read about the proposed network.

Graphic Abstract

Project dir

The implementation of PolarNet in different detection frameworks is identical:

• SOTA double-stage: Faster RCNN/ (based on the github repo AttFaster)
• SOTA one-stage: YOLOX/ (based on the github repo YOLOX)
• Others: mmdetection/ (based on the github repo mmdetection)

Data

Data is available except for images:

• COCO format annotation: data/mmdetection
• CSV format annotation: data/YOLOX & data/Faster RCNN

Details

Please see the corresponding README.md inside folders.