The code and the algorithm are for non-commercial use only.
Paper : "Automated Strabismus Detection for Telemedicine Applications"
Author: Jiewei Lu, Zhun Fan, Ce Zheng, Jingan Feng, Longtao Huang, Wenji Li, Erik D. Goodman
(12jwlu1@stu.edu.cn, zfan@stu.edu.cn, zhengce@hotmail.com, 13jafeng@stu.edu.cn, 17lthuang@stu.edu.cn, , liwj@stu.edu.cn, goodman@egr.msu.edu)
Date : December 2, 2018
Version : 2.0
Copyright (c) 2018, Jiewei Lu, Jingan Feng.
--------------------------------------------------------------
Notes:
1) Tensorflow, an open source machine learning framework, is required for the implementation.
2) R-FCN for eye region segmentation, is based on the TensorFlow Object Detection API. Please refer to the source code at https://github.com/tensorflow/models/tree/master/research/object_detection
3) Deep CNN for eye region classification, is based on the TensorFlow-Slim image classification model library. Please refer to the source code at https://github.com/tensorflow/models/tree/master/research/slim
4) scikit-learn, a machine learning library, is required for the evaluation metrics (ROC & AUC).
These libraries can be easily set by packet manager on linux systems
--------------------------------------------------------------
This folder contains two sub-directories:
- Eye_Region_Segmentation
- eye_detection.py the source code of using R-FCN to segment eye region
- iou.py the source code of displaying the mean IOU result of Segmentation
- IMG_FILE contains some example images for testing R-FCN
- XML_FILE contains corresponding bounding box information for each example image
- detection_result contains the numpy file saving IOU output of R-FCN for each images in test set
- Strabismus_Diagnosis
- eye_classification.py the source code of using deep CNN to classify eye region
- roc_auc.py the source code of calculating and displaying evaluation metrics of our deep CNN
- CROP_IMAGE contains some example images for testing the deep CNNs
- network contains the design file of the network architecture
- network_result contains the numpy file saving goundtruth labels and detected results of test set