Introduction

Comparative Test Unet & other variants

Depence

Python 3.5.7

Tensorflow 1.12.0

Pillow 6.0.0

numpy 1.15.2

opencv_python 4.1.0

Usage

To training a model and evaluate every epoch, create tfrecord for datas by using DataProcessor/LoadArguandSavetif.py.py , and put .tfrecord files in /Data . Run python main.py

To evaluate the trained model on the test set without ground-truth labels, create tfrecord for datas by using DataProcessor/LoadArguandSavetif.py.py , and put .tfrecord files in /Data . Run python Main.py --Mode=Visualize --mod_dir=[path to model] --img_dir=[path to test images].

Explanations

/Data/: Datas(Trainset.tfrecord Testset.tfrecord)

/DataProcessor/: Data processing tools LoadArguandSavetif.py: .tif to .tfrecord Processor.py : Data processing functions( Augmentation function, Image enhancement, Image Intensity, image splite&merge Image downsampling)

Readtf.py: Method for read train/test set

/Nets/: Network structure

Unet.py & Unet_Res.py& Unet_Res_Dia.py: Comparative test networks Layers.py & Layers_Dia.py: Layers packages

/TestImage/ & /TestRes/: Examples

Main.py: Main function

TFUtils.py & TestUtils.py: Train&Test Utils