def reconstruct_image(stacked_img, ori_shape, epoch, save_folder_name):
stacked_img = stacked_img.cpu().data.numpy()
stacked_img_shape = np.shape(stacked_img)
in_size = np.shape(stacked_img)[1:]
pdsz = 20
crop_n1, crop_n2, crop_n3 = cal_crop_num(ori_shape, in_size)
div_arr = division_array(stacked_img_shape[1:], crop_n1, crop_n2, crop_n3,
ori_shape) + 0.000001
img_cont_np = image_concatenate(stacked_img, crop_n1, crop_n2, crop_n3,
ori_shape[0], ori_shape[1], ori_shape[2])
img_cont = img_cont_np / div_arr
img_cont_np = img_cont.astype('float')
# pdsz = 30
img_cont_np = img_cont_np[pdsz:-pdsz, pdsz:-pdsz, pdsz:-pdsz]
# organize images in every epoch
desired_path = save_folder_name + '/epoch_' + str(epoch) + '/'
# Create the path if it does not exist
if not os.path.exists(desired_path):
os.makedirs(desired_path)
# Save Image!
export_name = str(epoch) + '_dis.nii.gz'
save_array_as_nii_volume(img_cont_np, desired_path + export_name)
return img_cont
def save_com_img2nii(pred, lab, save_path):
lab[lab == 255] = 5
pred[pred == 255] = 1
re = lab - pred
re[re == 255] = 3
re[re == 4] = 2
re[re == 5] = 1
print(np.unique(re[:]))
save_array_as_nii_volume(re, save_path)
def save_prediction_image(stacked_img,
ori_shape,
im_name,
epoch,
indate_it=0,
save_folder_name="result_images",
save_im=True):
"""save images to save_path
Args:
stacked_img (numpy): stacked cropped images
save_folder_name (str): saving folder name
division_array(388, 2, 3, 768, 1024):
"""
stacked_img = stacked_img.cpu().data.numpy()
stacked_img_shape = np.shape(stacked_img)
in_size = np.shape(stacked_img)[1:]
stacked_size = stacked_img_shape[0]
pdsz = 20
crop_n1, crop_n2, crop_n3 = cal_crop_num(ori_shape, in_size)
div_arr = division_array(stacked_img_shape[1:], crop_n1, crop_n2, crop_n3,
ori_shape) + 0.000001
img_cont_np = image_concatenate(stacked_img, crop_n1, crop_n2, crop_n3,
ori_shape[0], ori_shape[1], ori_shape[2])
probability = img_cont_np / div_arr
# print(np.unique(probability))
probability = probability[pdsz:-pdsz, pdsz:-pdsz, pdsz:-pdsz]
desired_path = save_folder_name + '/epoch_' + str(epoch) + '/'
# Create the path if it does not exist
if not os.path.exists(desired_path):
os.makedirs(desired_path)
# Save Image!
export_name = str(epoch) + '_pro.nii.gz'
save_array_as_nii_volume(probability, desired_path + export_name)
img_cont_np = img_cont_np.astype('uint8')
img_cont_np = polarize((img_cont_np) / div_arr) * 255
img_cont_np = img_cont_np[..., pdsz:-pdsz, pdsz:-pdsz, pdsz:-pdsz]
# organize images in every epoch
if indate_it == 0:
desired_path = save_folder_name + '/epoch_' + str(epoch) + '/'
else:
desired_path = save_folder_name + '/iter_' + str(epoch) + '/'
# Create the path if it does not exist
if not os.path.exists(desired_path):
os.makedirs(desired_path)
# Save Image!
export_name = str(epoch) + '.nii.gz'
save_array_as_nii_volume(img_cont_np, desired_path + export_name)
return img_cont_np
def save_img2nii(img_path, save_path):
img_arr = glob.glob(str(img_path) + str("/*"))
img_arr.sort()
single_image_name = img_arr[0]
img_as_img = Image.open(single_image_name)
img_as_np = np.asarray(img_as_img).astype('uint8')
all_lab = np.zeros((img_as_np.shape[0], img_as_np.shape[1], len(img_arr)))
for i in range(len(img_arr)):
single_image_name = img_arr[i]
img_as_img = Image.open(single_image_name)
img_as_np = np.asarray(img_as_img).astype('uint8')
all_lab[:, :, i] = img_as_np
# img_as_img.show()
save_array_as_nii_volume(np.transpose(all_lab, [2, 0, 1]), save_path)
import nibabel, glob
import numpy as np
from util.tools_self import save_array_as_nii_volume, load_nifty_volume_as_array
from PIL import Image
from metrics import *
from matplotlib import pyplot as plt
import scipy.misc
pdsz = 30
save_filename = 'train_yz_proximity.nii.gz'
img_path = 'oriCvLab/train_lab_c2_yz'
img_arr = glob.glob(str(img_path) + "/*")
img_arr.sort()
img = np.zeros((165, 768 // 2, 1024 // 2)).astype('uint8')
# lab = np.zeros((165,1024//2,768//2)).astype('uint8')
for i in range(len(img_arr)):
img_as_img = Image.open(img_arr[i])
# img_as_img.show()
img_as_np = np.asarray(img_as_img).astype('uint8')
# img_as_np = scipy.misc.imresize(img_as_np, 0.5)
img[:, :, i] = img_as_np
img = np.pad(img, pdsz, mode='symmetric')
save_array_as_nii_volume(np.transpose(img, [2, 0, 1]), save_filename)
print("pro1, dice is:", dice, "jac is:", jac)
pro2 = load_nifty_volume_as_array(flip_2_path)
# 165 * 512
pro2 = np.transpose(pro2, [1, 0, 2])
print(pro2.shape)
dice, jac = dice_coeff((pro2 >= 0.5).astype('uint8') * 255, target)
print("pro2, dice is:", dice, "jac is:", jac)
pro3 = load_nifty_volume_as_array(flip_3_path)
pro3 = np.transpose(pro3, [1, 2, 0])
print(pro3.shape)
dice, jac = dice_coeff((pro3 >= 0.5).astype('uint8') * 255, target)
print("pro3, dice is:", dice, "jac is:", jac)
desired_path = './history/'
pro = pro1 + pro2 + pro3
pro = pro / 3
export_name = 'aug_test_pro' + '.nii.gz'
save_array_as_nii_volume(pro, desired_path + export_name)
export_name = 'aug_test' + '.nii.gz'
pred = pro >= 0.5
pred = pred.astype('uint8')
save_array_as_nii_volume(pred, desired_path + export_name)
pred[pred == 1] = 255
print(np.unique(pred))
dice, jac = dice_coeff(pred, target)
print("dice is:", dice, "jac is:", jac)
save_path = './com.nii.gz'
save_com_img2nii(pred, target, save_path)