Exercise UNet

problem

I am trying to test my model with different batch sizes and I am getting different accuracies for different batch sizes.

The model has been roughly trained https://share.weiyun.com/ZhShMt5U. Modify the patchsize in test.py to test

structure

• data：The drishti data set contains 101 images and masks，all here, no more downloads.
• dataset.py：pytorch Dataset, and albumentations
• metrics.py：evaluating indicator like dice iou...
• net.py：unet
• process.py：Training function and verification function
• test.py：Training entrance
• train.py：Test entrance

Environment

• pytorch 1.11 (reference)
• torchvision
• numpy
• torch
• tqdm
• albumentations

Dataset introduce

Try to separate OC and OD from fundus image

run

cd  unet

pip install -r requirements.txt

python train.py

python test.py