from mnet_utils import dice_coef_loss, train_loader, mk_dir, return_list
import Model_MNet as DeepModel
result_path = mk_dir('./deep_model/')
pre_model_file = './deep_model/Model_MNet_REFUGE.h5'
save_model_file = result_path + 'Model_MNet_REFUGE_v2.h5'
root_path = './../training_crop/'
train_data_path = root_path + 'data/'
train_mask_path = root_path + 'label/'
val_data_path = root_path + 'val_data/data/'
val_mask_path = root_path + 'val_data/label/'
# load training data
train_list = return_list(train_data_path, '.png')
val_list = return_list(val_data_path, '.png')
Total_iter = 100
nb_epoch_setting = 3
input_size = 400
optimizer_setting = SGD(lr=0.0001, momentum=0.9)
my_model = DeepModel.DeepModel(size_set=input_size)
my_model.load_weights(pre_model_file, by_name=True)
my_model.compile(optimizer=optimizer_setting,
loss=dice_coef_loss,
loss_weights=[0.1, 0.1, 0.1, 0.1, 0.6])
import Model_DiscSeg as DiscModel
disc_list = [400, 500, 600, 700, 800]
DiscROI_size = 800
DiscSeg_size = 640
CDRSeg_size = 400
data_type = '.jpg'
data_img_path = '../data/REFUGE_Training400/Training400/Glaucoma/'
label_img_path = '../data/Annotation_Training400/Annotation_Training400/Disc_Cup_Masks/Glaucoma/'
data_save_path = mk_dir('../training_crop/data/')
label_save_path = mk_dir('../training_crop/label/')
file_test_list = return_list(data_img_path, data_type)
DiscSeg_model = DiscModel.DeepModel(size_set=DiscSeg_size)
DiscSeg_model.load_weights('./deep_model/Model_DiscSeg_ORIGA.h5')
for lineIdx in range(len(file_test_list)):
temp_txt = file_test_list[lineIdx]
print(' Processing Img ' + str(lineIdx + 1) + ': ' + temp_txt)
# load image
org_img = np.asarray(image.load_img(data_img_path + temp_txt))
# load label
org_label = np.asarray(
image.load_img(label_img_path + temp_txt[:-4] + '.bmp'))[:, :, 0]
from skimage.measure import label, regionprops
from time import time
import cv2
from mnet_utils import pro_process, BW_img, disc_crop, mk_dir, return_list
import Model_DiscSeg as DiscModel
import Model_MNet as MNetModel
DiscROI_size = 600
DiscSeg_size = 640
CDRSeg_size = 400
test_data_path = './test_img/'
data_save_path = mk_dir('./result/')
file_test_list = return_list(test_data_path, '.jpg')
DiscSeg_model = DiscModel.DeepModel(size_set=DiscSeg_size)
DiscSeg_model.load_weights('./deep_model/Model_DiscSeg_ORIGA.h5')
CDRSeg_model = MNetModel.DeepModel(size_set=CDRSeg_size)
CDRSeg_model.load_weights('./deep_model/Model_MNet_REFUGE.h5')
for lineIdx in range(len(file_test_list)):
temp_txt = file_test_list[lineIdx]
# load image
org_img = np.asarray(image.load_img(test_data_path + temp_txt))
# Disc region detection by U-Net
temp_img = resize(org_img, (DiscSeg_size, DiscSeg_size, 3))*255
temp_img = np.reshape(temp_img, (1,) + temp_img.shape)
import os
import numpy as np
import cv2
import matplotlib.pyplot as plt
from mnet_utils import BW_img, disc_crop, mk_dir, return_list
from Utils import *
test_data = 'UNetJSRT/val/IMG/'
test_mask = 'UNetJSRT/val/'
test_pred = 'UNetJSRT/test/'
root_path = 'UNetJSRT/'
test_files = return_list(test_data, '.png')
image = cv2.imread(test_data + test_files[19])
mask_image_right_lung = cv2.imread(test_mask + 'right_lung/' + test_files[19])
mask_image_left_lung = cv2.imread(test_mask + 'left_lung/' + test_files[19])
mask_image_heart = cv2.imread(test_mask + 'heart/' + test_files[19])
mask_image_left_clavicle = cv2.imread(test_mask + 'left_clavicle/' +
test_files[19])
mask_image_right_clavicle = cv2.imread(test_mask + 'right_clavicle/' +
test_files[19])
predict_mask_left_lung = cv2.imread(test_pred + 'class2/' + test_files[19])
predict_mask_right_lung = cv2.imread(test_pred + 'class1/' + test_files[19])
predict_mask_heart = cv2.imread(test_pred + 'class3/' + test_files[19])
predict_mask_left_clavicle = cv2.imread(test_pred + 'class4/' + test_files[19])
predict_mask_right_clavicle = cv2.imread(test_pred + 'class5/' +
test_files[19])
original_image = add_colored_mask_JSRT(image, mask_image_left_lung,
mask_image_right_lung, mask_image_heart,