def __inference__( self, index ):
image_path = self.images_path[index]
label_path = self.labels_path[index]
image = load_image(image_path, type=sitk.sitkFloat64)
label = load_image(label_path, type=sitk.sitkUInt8)
origin = label.GetOrigin()
spacing = label.GetSpacing()
direction = label.GetDirection()
image = truncation(image, lower=-200, upper=300)
image = sitk.GetArrayFromImage(image)
label = sitk.GetArrayFromImage(label)
# (W, H, D) -> (D, H, W)
image = np.transpose(image,axes=(2,1,0))
label = np.transpose(label,axes=(2,1,0))
shape = image.shape
image_patch, _, locations = sample(image, label, self.sample_shape, overlap=(0, 0, 0))
patch_result = []
for i in range(len(image_patch)):
patch = image_patch[i]
patch = patch[np.newaxis, np.newaxis, :]
patch = torch.from_numpy(patch)
patch = patch.float()
if self.use_gpu:
patch = patch.cuda()
with torch.no_grad():
output = self.model(patch)
patch_result.append(output.cpu().numpy()[0,:])
predict = assemble_channels(patch_result, shape, locations)
predict = np.argmax(predict, axis=0)
# (D, H, W) -> (W, H, D)
predict = np.transpose(predict, axes=(2,1,0))
predict = sitk.GetImageFromArray(predict)
predict.SetOrigin(origin)
predict.SetDirection(direction)
predict.SetSpacing(spacing)
inference_file_path = os.path.join(self.inference_dir, str(self.patient_ids[index]) + ".nii.gz")
sitk.WriteImage(predict, inference_file_path)
def __getitem__(self, index):
if self.skip_images:
img = None
pose = self.poses[index]
else:
if self.mode == 0:
img = None
while img is None:
img = load_image(self.c_imgs[index])
pose = self.poses[index]
index += 1
index -= 1
elif self.mode == 1:
img = None
while img is None:
img = load_image(self.d_imgs[index])
pose = self.poses[index]
index += 1
index -= 1
elif self.mode == 2:
c_img = None
d_img = None
while (c_img is None) or (d_img is None):
c_img = load_image(self.c_imgs[index])
d_img = load_image(self.d_imgs[index])
pose = self.poses[index]
index += 1
img = [c_img, d_img]
index -= 1
else:
raise Exception('Wrong mode {:d}'.format(self.mode))
if self.target_transform is not None:
pose = self.target_transform(pose)
if self.skip_images:
return img, pose
if self.transform is not None:
if self.mode == 2:
img = [self.transform(i) for i in img]
else:
img = self.transform(img)
return img, pose
def generate_batch(self):
start = self.start
end = self.end
batch_list = []
label_list = []
# cargamos las imagenes y estas son tratadas para darles el tamaño requerido
for i in range(start, end):
# print(self.images[i], i)
img = load_image(self.dir_images + str(self.images[i]) + self.type, scale=self.divisor_scale, dim_image=self.dim_image)[:, :, :3]
batch_list.append(img.reshape((1, self.dim_image, self.dim_image, 3)))
label_list.append(self.labels[i])
# concatena cada elemento del batch_list dando como resultado una matriz con la forma (n, 224, 224, 3)
# retorna el batch_list concatenado y el label_list con n items, una etiqueta para cada imagen
return np.concatenate(batch_list, 0), label_list
def __getitem__(self, index):
img = None
while img is None:
img = load_image(self.c_imgs[index])
tforms = [transforms.ToTensor()]
data_transform = transforms.Compose(tforms)
img1 = data_transform(img)
img2 = np.array(img1)
pose1 = self.poses[index]
index += 1
index -= 1
return img2,pose1
def __getitem__(self, index):
pose = self.poses[index]
if self.skip_images:
return None, pose
img = load_image(self.c_imgs[index])
#if True:
# img = img.convert('P')
if self.target_transform is not None:
pose = self.target_transform(pose)
if self.transform is not None:
if self.mode == 2:
img = [self.transform(i) for i in img]
else:
img = self.transform(img)
return img, pose
def generate_batch(self):
start = self.start
end = self.end
batch_list = []
label_list = []
# cargamos las imagenes y estas son tratadas para darles el tamaño requerido
for i in range(start, end):
lab = self.labels.iloc[i, :].values
img = utils.load_image(self.dir_images + str(self.images[i]) +
self.type)[:, :, :3]
batch_list.append(img.reshape((1, 224, 224, 3)))
label_list.append(lab)
# concatena cada elemento del batch_list dando como resultado una matriz con la forma (n, 224, 224, 3)
# retorna el batch_list concatenado y el label_list con n items, una etiqueta para cada imagen
return np.concatenate([block for block in batch_list], 0), label_list
def __getitem__(self, index):
pose = self.poses[index]
#print(pose)
if self.skip_images:
return None, pose
#print(self.c_imgs[index])
img = load_image(self.c_imgs[index])
#TODO: check aspect ratio
if img.size[0] > 720:
img = img.resize((720,480))
#print("Lodading image:", index,img.mode,img.size,self.c_imgs[index],pose)
if self.target_transform is not None:
pose = self.target_transform(pose)
#print(pose)
if self.transform is not None:
img = self.transform(img)
return img, pose
def write_data_hdf5(image_items,
mask_items,
hdf5_output_file,
process_filter_rate,
write_per_num=500,
process_resize_spacing=None,
sample_size=(128, 128, 32),
sample_overlap=(6, 6, 6),
window_lower=-200,
window_upper=300):
# setting-up HDF5 file
h5file = h5py.File(name=hdf5_output_file, mode='w')
data = {}
ds_shape = tuple([write_per_num] + list(sample_size))
max_shape = tuple([None] + list(sample_size))
data['image'] = h5file.create_dataset(
name='image',
shape=ds_shape,
compression='gzip',
dtype=np.float32,
maxshape=max_shape,
)
data['mask'] = h5file.create_dataset(
name='mask',
shape=ds_shape,
compression='gzip',
dtype=np.uint8,
maxshape=max_shape,
)
data['pid'] = h5file.create_dataset(
name='patient_id',
shape=tuple([write_per_num]),
compression='gzip',
dtype=h5py.special_dtype(vlen=np.str),
maxshape=tuple([None]),
)
data['location'] = h5file.create_dataset(name='location',
shape=tuple([write_per_num, 3]),
compression='gzip',
dtype=np.uint8,
maxshape=tuple([None, 3]))
# process one patient by one patient
index = 1
patch_num = 0
patch_writed = 0
image_patch_container = list()
mask_patch_container = list()
location_container = list()
pid_container = list()
for image_path, mask_path in zip(image_items, mask_items):
logging.info('-' * 100)
logging.info('start process %d patient' % index)
# step 1: load image
image = load_image(image_path, sitk.sitkFloat32)
mask = load_image(mask_path, sitk.sitkUInt8)
# step 2: truncate image with fixed window
image = truncation(image, lower=window_lower, upper=window_upper)
# step 3: resize z-axis
# set z-axis spacing to 1.0
original_spacing = image.GetSpacing()
if process_resize_spacing is 'standard':
resize_spacing = [1.0, 1.0, 1.0]
elif process_resize_spacing == 'origin':
resize_spacing = original_spacing
elif process_resize_spacing == 'z-axis':
resize_spacing = [1.0] + []
else:
raise NotImplementedError(
'Not support for this resize spacing, please check again.')
image = resize(image,
resize_spacing,
original_spacing,
sitk.sitkLinear,
dtype=sitk.sitkFloat32)
mask = resize(mask,
resize_spacing,
original_spacing,
sitk.sitkUInt8,
dtype=sitk.sitkUInt8)
image = sitk.GetArrayFromImage(image)
mask = sitk.GetArrayFromImage(mask)
# step 4: sample patch
image_patchs, mask_patchs, locations = sample(image,
mask,
size=sample_size,
overlap=sample_overlap)
logging.info('sample from {} with {} patchs'.format(
image_path, str(len(image_patchs))))
# step 5: filter patch
image_patchs, mask_patchs, locations = patch_filter(
image_patchs,
mask_patchs,
background=0,
rate=process_filter_rate,
locations=locations)
logging.info('filter form {} with {} patchs'.format(
image_path, str(len(image_patchs))))
length = len(image_patchs)
if length == 0:
index += 1
continue
image_patch_container.extend(image_patchs)
mask_patch_container.extend(mask_patchs)
location_container.extend(locations)
pid_container.extend(len(image_patchs) * [image_path])
if len(image_patch_container) >= write_per_num:
for i in range(len(image_patch_container) // write_per_num):
image_patch_to_save = np.stack(
image_patch_container[:write_per_num])
mask_patch_to_save = np.stack(
mask_patch_container[:write_per_num])
location_to_save = np.stack(
mask_patch_container[:write_per_num])
pid_to_save = np.stack(pid_container[:write_per_num])
data['image'][patch_writed:patch_writed +
write_per_num, :] = image_patch_to_save
data['mask'][patch_writed:patch_writed +
write_per_num, :] = mask_patch_to_save
data['location'][patch_writed:patch_writed +
write_per_num, :] = location_to_save
data['pid'][patch_writed:patch_writed +
write_per_num, :] = pid_to_save
patch_writed += write_per_num
image_patch_container = image_patch_container[write_per_num:]
mask_patch_container = mask_patch_container[write_per_num:]
location_container = location_container[write_per_num:]
pid_container = pid_container[write_per_num:]
patch_num += length
index += 1
if len(image_patch_container) != 0:
length = len(image_patch_container)
image_patch_to_save = np.stack(image_patch_container)
mask_patch_to_save = np.stack(mask_patch_container)
location_to_save = np.stack(location_container)
pid_to_save = np.stack(pid_container)
data['image'].resize(size=patch_writed + length, axis=0)
data['mask'].resize(size=patch_writed + length, axis=0)
data['pid'].resize(size=patch_writed + length, axis=0)
data['location'].resize(size=patch_writed + length, axis=0)
data['image'][patch_writed:patch_writed +
length, :] = image_patch_to_save
data['mask'][patch_writed:patch_writed +
length, :] = mask_patch_to_save
data['pid'][patch_writed:patch_writed + length] = pid_to_save
data['location'][patch_writed:patch_writed +
length, :] = location_to_save
patch_writed += length
logging.info('-' * 100)
logging.info('Preprocess complete, total processed %d patchs' %
(patch_num))
h5file.close()
import numpy as np
import tensorflow as tf
import os, sys
PATH_DIRECTORY = os.path.abspath('../..')
PATH_TESTER_DIR = os.path.dirname('__file__')
PATH_SRC_DIR = '..'
sys.path.insert(0, os.path.abspath(os.path.join(PATH_TESTER_DIR,
PATH_SRC_DIR)))
from tools.utils import load_image, show_image, print_prob_all
from nets.vgg19 import cnn_vgg19
img1 = load_image(PATH_DIRECTORY + '/data/test/tigerTrain1.jpeg')
img2 = load_image(PATH_DIRECTORY + '/data/test/tigerTrain2.jpeg')
img3 = load_image(PATH_DIRECTORY + '/data/test/tigerTrain3.jpg')
img4 = load_image(PATH_DIRECTORY + '/data/test/tigerTest1.jpg')
img5 = load_image(PATH_DIRECTORY + '/data/test/tigerTest2.jpg')
batch1 = img1.reshape((1, 224, 224, 3))
batch2 = img2.reshape((1, 224, 224, 3))
batch3 = img3.reshape((1, 224, 224, 3))
batch4 = img4.reshape((1, 224, 224, 3))
batch5 = img5.reshape((1, 224, 224, 3))
batch_train = np.concatenate((batch1, batch2, batch3), 0)
label = [292, 292, 292]
batch_test = np.concatenate((batch4, batch5), 0)
batch = img1.reshape((1, 224, 224, 3))
import numpy as np
import tensorflow as tf
import os, sys
PATH_DIRECTORY = os.path.abspath('../..')
PATH_TESTER_DIR = os.path.dirname('__file__')
PATH_SRC_DIR = '..'
sys.path.insert(0, os.path.abspath(os.path.join(PATH_TESTER_DIR,
PATH_SRC_DIR)))
from tools.utils import load_image, show_image, print_prob_all
from nets.vgg19 import cnn_vgg19
img1 = load_image(PATH_DIRECTORY + '/data/example/avion.jpeg')
img2 = load_image(PATH_DIRECTORY + '/data/example/tiger.jpeg')
batch1 = img1.reshape((1, 224, 224, 3))
batch2 = img2.reshape((1, 224, 224, 3))
batch = np.concatenate((batch1, batch2), 0)
npy_path = PATH_DIRECTORY + '/weights/vgg19/vgg19.npy'
data_label_path = PATH_DIRECTORY + '/data/synset.txt'
with tf.Session() as sess:
# VARIABLES
images = tf.placeholder(tf.float32, [2, 224, 224, 3])
train_mode = tf.placeholder(tf.bool)
# MODEL VGG19
vgg19 = cnn_vgg19(npy_path, trainable=False)
vgg19.build(images, train_mode=train_mode)
