Glaucoma Master

---

Glaucoma Diagnosis model

1. meta

• Python 3.6.8
• conda 4.6.14
• tensorflow-gpu 1.13.1
• CUDA 9.0
• NVIDIA cuDNN v7.5.0 (Feb 21, 2019), for CUDA 9.0
• opencv 3.4.3

Train segmentation

train segmentation model

python train_seg.py

you can test source code with -t option (it will load small amount of data for testing)

python train_seg.py -t

test segmentation

visualize segmentation image for vaildation

test.ipynb

Generate segmentation image

generate segmentation image by using trained model

python mask_image_generator.py -m 1
python mask_image_generator.py -m 0

• options
  • 1: train image
  • 0: test image

generate train/test dataset

it will create hdf5 dataset, dataset will store preprocessed image for speed up training

python binary_dump.py
python binary_dump.py -t # test mode
python binary_dump.py -c # copy mode

• "copy mode" will copy Glaucoma image to balance Glaucoma and Non-Glaucoma image

Train model

train Diagnosis model

python train.py

Result

you can verify model performance in jupyter notebook

result.ipynb