Retina Image Segmentation with UNet

2018-06-29

1.说明

基于UNet的眼底图像血管分割方法：Retina-UNet，项目地址：https://github.com/orobix/retina-unet

2.配置运行环境及依赖：

Ubuntu-16 + CUDA-9.0 + cudnn-7.0 + GPUDriver-390.67 + OpenCV-3.2

Python 2.7
TensorFlow 1.8.0
TensorFlow-GPU 1.8.0 #不安装，则直接用CPU训练
Keras 2.2.0 #原release使用的1.1.0版本keras，若使用keras2.x，则需要修改部分代码
PIL (pillow 3.1.2)
h5py 2.8.0
ConfigParser 3.5.0
Numpy 1.14.5
scikit-learn 0.19.1
pydot 1.2.4

$ pip install xxx --user #安装上面这些依赖项

GraphViz

$ sudo apt-get install graphviz libgraphviz-dev

3.数据集

训练用的是视网膜血管分割经典的数据集DRIVE：http://www.isi.uu.nl/Research/Databases/DRIVE/，其中包含手工分割的ground-truth和mask图像。

下载解压后放于项目根目录下：

DRIVE
│
└───test
|    ├───1st_manual
|    └───2nd_manual
|    └───images
|    └───mask
│
└───training
    ├───1st_manual
    └───images
    └───mask

4.训练

首先整理数据：

python prepare_datasets_DRIVE.py

执行训练：

配置文件可修改configuration.txt

pytohn run_train.py


1. Create directory for the results (if not already existing)
Dir already existing
copy the configuration file in the results folder
2. Run the training on GPU (no nohup)
Using TensorFlow backend.
train images/masks shape:
(20, 1, 565, 565)
train images range (min-max): 0.0 - 1.0
train masks are within 0-1
patches per full image: 9500
train PATCHES images/masks shape:
(190000, 1, 48, 48)
train PATCHES images range (min-max): 0.0 - 1.0
2018-06-30 00:43:12.163406: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA
2018-06-30 00:43:12.584228: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:898] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2018-06-30 00:43:12.585652: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1356] Found device 0 with properties: 
name: GeForce GTX TITAN X major: 5 minor: 2 memoryClockRate(GHz): 1.076
pciBusID: 0000:01:00.0
totalMemory: 11.92GiB freeMemory: 11.57GiB
2018-06-30 00:43:12.585718: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1435] Adding visible gpu devices: 0
2018-06-30 00:43:20.161626: I tensorflow/core/common_runtime/gpu/gpu_device.cc:923] Device interconnect StreamExecutor with strength 1 edge matrix:
2018-06-30 00:43:20.161703: I tensorflow/core/common_runtime/gpu/gpu_device.cc:929]      0 
2018-06-30 00:43:20.161729: I tensorflow/core/common_runtime/gpu/gpu_device.cc:942] 0:   N 
2018-06-30 00:43:20.174512: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1053] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 11206 MB memory) -> physical GPU (device: 0, name: GeForce GTX TITAN X, pci bus id: 0000:01:00.0, compute capability: 5.2)
Check: final output of the network:
(None, 2304, 2)
Train on 171000 samples, validate on 19000 samples
Epoch 1/150
 - 204s - loss: 0.2764 - acc: 0.9045 - val_loss: 0.2325 - val_acc: 0.9240
Epoch 00001: val_loss improved from inf to 0.23253, saving model to ./train/train_best_weights.h5
Epoch 2/150
 - 199s - loss: 0.1679 - acc: 0.9380 - val_loss: 0.1698 - val_acc: 0.9420
Epoch 00002: val_loss improved from 0.23253 to 0.16979, saving model to ./train/train_best_weights.h5
Epoch 3/150

5.测试

result

6.Debug：

1) Keras2版本不兼容的问题：

from keras.utils.visualize_util import plot
ImportError: No module named 'keras.utils.visualize_util'

From visualize_util to vis_utils;
The plot function was also renamed to plot_model;
The nb_epoch argument in fit has been renamed epochs;
Update your Model call to the Keras 2 API: Model(outputs=Tensor("ac..., inputs=Tensor("in...)
model = Model(input=inputs, output=conv7);
更新代码或者选用合适的Keras版本。

2) ImportError: No module named tensorflow
使用pip install tensorflow安装失败且出现importing error，解决方法是在命令后加上--user。

3) ImportError: Failed to import pydot. Please install pydot. For example with pip install pydot.

4) pydot failed to call GraphViz. Please install GraphViz https://www.graphviz.org/ and ensure that its executables are in the $PATH. 在~/.bashrc中添加了PATH后错误仍然出现，不知道问题在哪。。。暂时注释掉retinaNN_training.py中的#plot_model(model, to_file='./'+name_experiment+'/'+name_experiment + '_model.png') #check how the model looks like。