/drusen_seg

The drusen segmentation model for early AMD.

Primary LanguageJupyter Notebook

The drusen segmentation model for early AMD.

Prerequisites

  • Python 2.7 or Python 3.3+
  • Pytorch 1.2.0
  • sklearn
  • albumentations
  • cv2

Testing

  • Step 1: predict the segmentation mask with the trained DeepLab model. Please check the notebook file test_segmentation_Deep_lab.ipynb (Change all the PATH director before runing the notebook.)
  • Step 2: predict the final drusen mask based on the results of step 1. Please check the notebook file test_main_model.ipynb (Change all the PATH director before runing the notebook.)

Note: the pre-trained models are provided in this link.

Training

  • Step 1: train the DeepLab model. Please check the notebook file train_segmentation_Deep_lab.ipynb (Change all the PATH director before runing the notebook.)
  • Step 2: predict the segmentation mask with the trained DeepLab model from step 1. Please check the notebook file test_segmentation_Deep_lab.ipynb (Change all the PATH director before runing the notebook.)
  • Step 3: Crop images, ground truth drusen mask and prediction drusen mask from trained DeepLab model (step 2). Please check the notebook file patch_extract.ipynb (Change all the PATH director before runing the notebook.)
  • Step 4: train the main model. Please check the notebook file train_main_model.ipynb (Change all the PATH director before runing the notebook.)

Results

alt text