thesuperorange / diabetic_aimarket

Diabetic_AImarket

Introduction

• 目標：透過有標記之眼底影像來偵測是否有糖尿病引發的視網膜病變
• 類別：分為五種嚴重程度
  • 0 - No DR
  • 1 - Mild
  • 2 - Moderate
  • 3 - Severe
  • 4 - Proliferative DR
• framework: Pytorch

Data preparation

• download data.zip

• For training

  • Text data
    • train_img.csv / train_img_small.csv
    • train_label.csv/ train_label_small.csv
    • test_img.csv/ test_img_small.csv
    • test_label.csv/ test_label_small.csv
  • Image data
    • data/<#>_left.jpeg
    • data/<#>_right.jpeg

資料展示

• 1.fundus_data.ipynb

• 內容--資料觀察

  1. label distribution

  2. preview images

     

訓練模型

• 2.DR_main.ipynb

• 內容

  1. Dataloader

  2. Training

  3. Plot results

     • accuracy comparison

       

     • confusion matrix

       

模型佈署

• 3.deploy_model.ipynb

Reference

• From Kaggle: https://www.kaggle.com/c/diabetic-retinopathy-detection#description