def test_all_nets():
data = 2
data_path = '/exports/lkeb-hpc/syousefi/Data/invivo_decomposed/1/'
# parent_path = '/exports/lkeb-hpc/syousefi/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/sythesize_code/ASL_LOG/normal_SSIM_p_a_newdb/experiment-1/'
parent_path = '/exports/lkeb-hpc/syousefi/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/sythesize_code/ASL_LOG/Workshop_log/MSE/experiment-2/'
# data_path = '/srv/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/Data-01/BrainWeb_permutation00_low/'
img_name = ''
label_name = ''
result_dir = 'results_mse/'
_rd = _invivo_read_data(data=data,
reverse=False,
img_name=img_name,
label_name=label_name,
dataset_path=data_path)
'''read path of the images for train, test, and validation'''
test_data = _rd.read_data_path()
chckpnt_dir = parent_path + 'unet_checkpoints/'
result_path = parent_path + result_dir
batch_no = 1
batch_no_validation = batch_no
# label_patchs_size = 87#39 # 63
# patch_window = 103#53 # 77#89
img_row1 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row2 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row3 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row4 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row5 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row6 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row7 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row8 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
mri_ph = tf.placeholder(tf.float32,
shape=[None, None, None, None, 1],
name='mri')
segmentation = tf.placeholder(tf.float32,
shape=[None, None, None, None, 1],
name='segmentation')
label1 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label2 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label3 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label4 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label5 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label6 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label7 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label8 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label9 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label10 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label11 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label12 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label13 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label14 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
all_loss = tf.placeholder(tf.float32, name='loss')
is_training = tf.placeholder(tf.bool, name='is_training')
input_dim = tf.placeholder(tf.int32, name='input_dim')
# ave_huber = tf.placeholder(tf.float32, name='huber')
densenet = _densenet()
y, degree = densenet.densenet(img_row1=img_row1,
img_row2=img_row2,
img_row3=img_row3,
img_row4=img_row4,
img_row5=img_row5,
img_row6=img_row6,
img_row7=img_row7,
img_row8=img_row8,
input_dim=input_dim,
mri=mri_ph,
is_training=is_training)
loss_instance = _loss_func()
labels = []
labels.append(label1)
labels.append(label2)
labels.append(label3)
labels.append(label4)
labels.append(label5)
labels.append(label6)
labels.append(label7)
labels.append(label8)
labels.append(label9)
labels.append(label10)
labels.append(label11)
labels.append(label12)
labels.append(label13)
labels.append(label14)
logits = []
logits.append(y[:, :, :, :, 0, np.newaxis])
logits.append(y[:, :, :, :, 1, np.newaxis])
logits.append(y[:, :, :, :, 2, np.newaxis])
logits.append(y[:, :, :, :, 3, np.newaxis])
logits.append(y[:, :, :, :, 4, np.newaxis])
logits.append(y[:, :, :, :, 5, np.newaxis])
logits.append(y[:, :, :, :, 6, np.newaxis])
logits.append(y[:, :, :, :, 7, np.newaxis])
logits.append(y[:, :, :, :, 8, np.newaxis])
logits.append(y[:, :, :, :, 9, np.newaxis])
logits.append(y[:, :, :, :, 10, np.newaxis])
logits.append(y[:, :, :, :, 11, np.newaxis])
logits.append(y[:, :, :, :, 12, np.newaxis])
logits.append(y[:, :, :, :, 13, np.newaxis])
with tf.name_scope('Loss'):
loss_dic = loss_instance.loss_selector('SSIM_perf_angio',
labels=labels,
logits=logits,
angio_SSIM_weight=1)
cost = tf.reduce_mean(loss_dic["loss"], name="cost")
# cost_angio = tf.reduce_mean(loss_dic["angio_SSIM"], name="angio_SSIM")
# cost_perf = tf.reduce_mean(loss_dic["perf_SSIM"], name="perf_SSIM")
# ========================================================================
# ave_loss = tf.placeholder(tf.float32, name='loss')
# ave_loss_perf = tf.placeholder(tf.float32, name='loss_perf')
# ave_loss_angio = tf.placeholder(tf.float32, name='loss_angio')
#
# average_gradient_perf = tf.placeholder(tf.float32, name='grad_ave_perf')
# average_gradient_angio = tf.placeholder(tf.float32, name='grad_ave_angio')
#
# ave_huber = tf.placeholder(tf.float32, name='huber')
# restore the model
sess = tf.Session()
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(chckpnt_dir)
saver.restore(sess, ckpt.model_checkpoint_path)
# copyfile('./test_synthesize_ssim_perf_angio.py', result_path + '/test_synthesize_ssim_perf_angio.py')
_image_class = invivo_image_class(test_data,
bunch_of_images_no=1,
is_training=1,
patch_window=patch_window,
sample_no_per_bunch=1,
label_patch_size=label_patchs_size,
validation_total_sample=0)
learning_rate = 1E-5
extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(extra_update_ops):
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost)
# init = tf.global_variables_initializer()
dic_perf0 = []
dic_perf1 = []
dic_perf2 = []
dic_perf3 = []
dic_perf4 = []
dic_perf5 = []
dic_perf6 = []
dic_angio0 = []
dic_angio1 = []
dic_angio2 = []
dic_angio3 = []
dic_angio4 = []
dic_angio5 = []
dic_angio6 = []
loss = 0
#min ASL crush : 0.499999, max ASL crush : 0.500000
#min ASL noncrush : 0.499984, max ASL noncrush : 0.500000
#min perfusion : 2.192158, max perfusion : 11.602790
#min angio : 0.000000, max angio : 1309.346558
max_angio = 1309.346558
max_perf = 11.602790
for img_indx in range(len(test_data)):
crush, noncrush, perf, angio, mri, segmentation_, spacing, direction, origin = _image_class.read_image_for_test(
test_set=test_data,
img_indx=img_indx,
input_size=in_dim,
final_layer=final_layer,
invivo=1)
for i in range(8):
crush[i][np.where(crush[i] < 0)] = 0
# crush[i]=np.flipud(crush[i])
crush = crush / np.max(crush)
crush = crush * (0.500000)
for i in range(8):
noncrush[i][np.where(noncrush[i] < 0)] = 0
# noncrush[i] = np.flipud(noncrush[i])
noncrush = noncrush / np.max(noncrush)
noncrush = noncrush * (0.500000)
for i in range(7):
perf[i][np.where(perf[i] < 0)] = 0
# perf[i] = np.flipud(perf[i])
perf = perf / np.max(perf) * max_perf
for i in range(7):
angio[i][np.where(angio[i] < 0)] = 0
# angio[i] = np.flipud(angio[i])
angio = angio / np.max(angio) * max_angio
[out] = sess.run(
[y],
feed_dict={
img_row1: np.expand_dims(np.expand_dims(crush[0], 0), -1),
img_row2: np.expand_dims(np.expand_dims(noncrush[1], 0), -1),
img_row3: np.expand_dims(np.expand_dims(crush[2], 0), -1),
img_row4: np.expand_dims(np.expand_dims(noncrush[3], 0), -1),
img_row5: np.expand_dims(np.expand_dims(crush[4], 0), -1),
img_row6: np.expand_dims(np.expand_dims(noncrush[5], 0), -1),
img_row7: np.expand_dims(np.expand_dims(crush[6], 0), -1),
img_row8: np.expand_dims(np.expand_dims(noncrush[7], 0), -1),
mri_ph: np.expand_dims(np.expand_dims(crush[0], 0),
-1), #useless input
segmentation: np.expand_dims(np.expand_dims(crush[0], 0),
-1), #useless input
label1: np.expand_dims(np.expand_dims(perf[0], 0), -1),
label2: np.expand_dims(np.expand_dims(perf[1], 0), -1),
label3: np.expand_dims(np.expand_dims(perf[2], 0), -1),
label4: np.expand_dims(np.expand_dims(perf[3], 0), -1),
label5: np.expand_dims(np.expand_dims(perf[4], 0), -1),
label6: np.expand_dims(np.expand_dims(perf[5], 0), -1),
label7: np.expand_dims(np.expand_dims(perf[6], 0), -1),
label8: np.expand_dims(np.expand_dims(angio[0], 0), -1),
label9: np.expand_dims(np.expand_dims(angio[1], 0), -1),
label10: np.expand_dims(np.expand_dims(angio[2], 0), -1),
label11: np.expand_dims(np.expand_dims(angio[3], 0), -1),
label12: np.expand_dims(np.expand_dims(angio[4], 0), -1),
label13: np.expand_dims(np.expand_dims(angio[5], 0), -1),
label14: np.expand_dims(np.expand_dims(angio[6], 0), -1),
is_training: False,
input_dim: patch_window,
all_loss: -1.,
})
parent_path = '/exports/lkeb-hpc/syousefi/Data/invivo_decomposed/1/00_PP0_0_cinema1/'
# result_dir='normal_SSIM/'
result_dir = 'results_mse/'
for i in range(np.shape(out)[-1]):
image = out[0, :, :, :, i]
sitk_image = sitk.GetImageFromArray(image)
res_dir = test_data[img_indx][0][0].split('/')[-2]
if i == 0:
os.mkdir(parent_path + result_dir + res_dir)
if i < 7:
nm = 'perf'
else:
nm = 'angi'
sitk_image.SetDirection(direction=direction)
sitk_image.SetOrigin(origin=origin)
sitk_image.SetSpacing(spacing=spacing)
sitk.WriteImage(
sitk_image, parent_path + result_dir + res_dir + '/' + nm +
'_' + str(i % 7) + '.nii')
print(parent_path + result_dir + res_dir + '/' + nm + '_' +
str(i % 7) + '.nii done!')
for i in range(7):
if i == 0:
os.mkdir(parent_path + result_dir + res_dir + '/GT/')
sitk_angio = sitk.GetImageFromArray(angio[i])
sitk_angio.SetDirection(direction=direction)
sitk_angio.SetOrigin(origin=origin)
sitk_angio.SetSpacing(spacing=spacing)
sitk.WriteImage(
sitk_angio, parent_path + result_dir + res_dir + '/GT/angio_' +
str(i) + '.nii')
sitk_perf = sitk.GetImageFromArray(perf[i])
sitk_perf.SetDirection(direction=direction)
sitk_perf.SetOrigin(origin=origin)
sitk_perf.SetSpacing(spacing=spacing)
sitk.WriteImage(
sitk_perf, parent_path + result_dir + res_dir + '/GT/perf_' +
str(i) + '.nii')
dic_perf0.append(
anly.analysis(out[0, :, :, :, 0], perf[i], 0, max_perf))
dic_perf1.append(
anly.analysis(out[0, :, :, :, 1], perf[i], 0, max_perf))
dic_perf2.append(
anly.analysis(out[0, :, :, :, 2], perf[i], 0, max_perf))
dic_perf3.append(
anly.analysis(out[0, :, :, :, 3], perf[i], 0, max_perf))
dic_perf4.append(
anly.analysis(out[0, :, :, :, 4], perf[i], 0, max_perf))
dic_perf5.append(
anly.analysis(out[0, :, :, :, 5], perf[i], 0, max_perf))
dic_perf6.append(
anly.analysis(out[0, :, :, :, 6], perf[i], 0, max_perf))
dic_angio0.append(
anly.analysis(out[0, :, :, :, 7], angio[i], 0, max_angio))
dic_angio1.append(
anly.analysis(out[0, :, :, :, 8], angio[i], 0, max_angio))
dic_angio2.append(
anly.analysis(out[0, :, :, :, 9], angio[i], 0, max_angio))
dic_angio3.append(
anly.analysis(out[0, :, :, :, 10], angio[i], 0, max_angio))
dic_angio4.append(
anly.analysis(out[0, :, :, :, 11], angio[i], 0, max_angio))
dic_angio5.append(
anly.analysis(out[0, :, :, :, 12], angio[i], 0, max_angio))
dic_angio6.append(
anly.analysis(out[0, :, :, :, 13], angio[i], 0, max_angio))
if img_indx == 0:
headers = dic_perf0[0].keys()
dics = [
dic_perf0, dic_perf1, dic_perf2, dic_perf3, dic_perf4, dic_perf5,
dic_perf6, dic_angio0, dic_angio1, dic_angio2, dic_angio3,
dic_angio4, dic_angio5, dic_angio6
]
save_in_xlsx(parent_path, headers, dics=dics)
print('Total loss: ', loss / len(test_data))
def test_all_nets():
data = 2
Server = 'DL'
if Server == 'DL':
parent_path = '/srv/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/sythesize_code/ASL_LOG/multi_stage/experiment-2/'
data_path = '/srv/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/Data-01/BrainWeb_permutation00_low/'
else:
parent_path = '/exports/lkeb-hpc/syousefi/Code/'
data_path = '/srv/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/Data-01/BrainWeb_permutation00_low/'
img_name = ''
label_name = ''
_rd = _read_data(data=data,
reverse=False,
img_name=img_name,
label_name=label_name,
dataset_path=data_path)
'''read path of the images for train, test, and validation'''
train_data, validation_data, test_data = _rd.read_data_path()
chckpnt_dir = parent_path + 'unet_checkpoints/'
result_path = parent_path + 'results/'
batch_no = 1
batch_no_validation = batch_no
# label_patchs_size = 87#39 # 63
# patch_window = 103#53 # 77#89
if test_vali == 1:
test_set = validation_data
else:
test_set = test_data
# ===================================================================================
# img_row1 = tf.placeholder(tf.float32,
# shape=[batch_no, patch_window, patch_window, patch_window, 1],
# name='img_row1')
# img_row2 = tf.placeholder(tf.float32,
# shape=[batch_no, patch_window, patch_window, patch_window, 1],
# name='img_row2')
# img_row3 = tf.placeholder(tf.float32,
# shape=[batch_no, patch_window, patch_window, patch_window, 1],
# name='img_row3')
# img_row4 = tf.placeholder(tf.float32,
# shape=[batch_no, patch_window, patch_window, patch_window, 1],
# name='img_row4')
# img_row5 = tf.placeholder(tf.float32,
# shape=[batch_no, patch_window, patch_window, patch_window, 1],
# name='img_row5')
# img_row6 = tf.placeholder(tf.float32,
# shape=[batch_no, patch_window, patch_window, patch_window, 1],
# name='img_row6')
# img_row7 = tf.placeholder(tf.float32,
# shape=[batch_no, patch_window, patch_window, patch_window, 1],
# name='img_row7')
# img_row8 = tf.placeholder(tf.float32,
# shape=[batch_no, patch_window, patch_window, patch_window, 1],
# name='img_row8')
#
# mri_ph = tf.placeholder(tf.float32,
# shape=[batch_no, patch_window, patch_window, patch_window, 1],
# name='mri')
#
# segmentation = tf.placeholder(tf.float32,
# shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='segments')
#
# label1 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label1')
# label2 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label2')
# label3 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label3')
# label4 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label4')
# label5 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label5')
# label6 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label6')
# label7 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label7')
# label8 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label8')
# label9 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label9')
# label10 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label10')
# label11 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label11')
# label12 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label12')
# label13 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label13')
# label14 = tf.placeholder(tf.float32, shape=[batch_no, label_patchs_size, label_patchs_size,
# label_patchs_size, 1], name='label14')
img_row1 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row2 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row3 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row4 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row5 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row6 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row7 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row8 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
mri_ph = tf.placeholder(tf.float32,
shape=[None, None, None, None, 1],
name='mri')
# segmentation = tf.placeholder(tf.float32, shape=[None, None, None, None, 1], name='segmentation')
label1 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label2 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label3 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label4 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label5 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label6 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label7 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label8 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label9 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label10 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label11 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label12 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label13 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label14 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
all_loss = tf.placeholder(tf.float32, name='loss')
is_training = tf.placeholder(tf.bool, name='is_training')
input_dim = tf.placeholder(tf.int32, name='input_dim')
# ave_huber = tf.placeholder(tf.float32, name='huber')
multi_stage_densenet = _multi_stage_densenet()
y, loss_upsampling11, loss_upsampling22 = multi_stage_densenet.multi_stage_densenet(
img_row1=img_row1,
img_row2=img_row2,
img_row3=img_row3,
img_row4=img_row4,
img_row5=img_row5,
img_row6=img_row6,
img_row7=img_row7,
img_row8=img_row8,
input_dim=input_dim,
mri=mri_ph,
is_training=is_training)
loss_instance = _loss_func()
labels = []
labels.append(label1)
labels.append(label2)
labels.append(label3)
labels.append(label4)
labels.append(label5)
labels.append(label6)
labels.append(label7)
labels.append(label8)
labels.append(label9)
labels.append(label10)
labels.append(label11)
labels.append(label12)
labels.append(label13)
labels.append(label14)
logits = []
logits.append(y[:, :, :, :, 0, np.newaxis])
logits.append(y[:, :, :, :, 1, np.newaxis])
logits.append(y[:, :, :, :, 2, np.newaxis])
logits.append(y[:, :, :, :, 3, np.newaxis])
logits.append(y[:, :, :, :, 4, np.newaxis])
logits.append(y[:, :, :, :, 5, np.newaxis])
logits.append(y[:, :, :, :, 6, np.newaxis])
logits.append(y[:, :, :, :, 7, np.newaxis])
logits.append(y[:, :, :, :, 8, np.newaxis])
logits.append(y[:, :, :, :, 9, np.newaxis])
logits.append(y[:, :, :, :, 10, np.newaxis])
logits.append(y[:, :, :, :, 11, np.newaxis])
logits.append(y[:, :, :, :, 12, np.newaxis])
logits.append(y[:, :, :, :, 13, np.newaxis])
stage1 = []
stage1.append(loss_upsampling11[:, :, :, :, 0, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 1, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 2, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 3, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 4, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 5, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 6, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 7, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 8, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 9, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 10, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 11, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 12, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 13, np.newaxis])
stage2 = []
stage2.append(loss_upsampling22[:, :, :, :, 0, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 1, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 2, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 3, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 4, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 5, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 6, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 7, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 8, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 9, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 10, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 11, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 12, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 13, np.newaxis])
with tf.name_scope('Loss'):
loss_dic = loss_instance.loss_selector(
'Multistage_ssim_perf_angio_loss',
labels=labels,
logits=logits,
stage1=stage1,
stage2=stage2)
cost = tf.reduce_mean(loss_dic["loss"], name="cost")
# cost_angio = tf.reduce_mean(loss_dic["angio_SSIM"], name="angio_SSIM")
# cost_perf = tf.reduce_mean(loss_dic["perf_SSIM"], name="perf_SSIM")
# ========================================================================
# ave_loss = tf.placeholder(tf.float32, name='loss')
# ave_loss_perf = tf.placeholder(tf.float32, name='loss_perf')
# ave_loss_angio = tf.placeholder(tf.float32, name='loss_angio')
#
# average_gradient_perf = tf.placeholder(tf.float32, name='grad_ave_perf')
# average_gradient_angio = tf.placeholder(tf.float32, name='grad_ave_angio')
#
# ave_huber = tf.placeholder(tf.float32, name='huber')
# restore the model
sess = tf.Session()
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(chckpnt_dir)
saver.restore(sess, ckpt.model_checkpoint_path)
copyfile('./test_synthesize_multistage_perf_angio.py',
result_path + '/test_synthesize_multistage_perf_angio.py')
_image_class = image_class(train_data,
bunch_of_images_no=1,
is_training=1,
patch_window=patch_window,
sample_no_per_bunch=1,
label_patch_size=label_patchs_size,
validation_total_sample=0)
learning_rate = 1E-5
extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(extra_update_ops):
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost)
# init = tf.global_variables_initializer()
dic_perf0 = []
dic_perf1 = []
dic_perf2 = []
dic_perf3 = []
dic_perf4 = []
dic_perf5 = []
dic_perf6 = []
dic_angio0 = []
dic_angio1 = []
dic_angio2 = []
dic_angio3 = []
dic_angio4 = []
dic_angio5 = []
dic_angio6 = []
loss = 0
Elapsed = []
for img_indx in range(2): #len(test_set)):
crush, noncrush, perf, angio, mri, segmentation_, spacing, direction, origin = _image_class.read_image_for_test(
test_set=test_set,
img_indx=img_indx,
input_size=in_dim,
final_layer=final_layer)
t = time.time()
[out] = sess.run(
[y],
feed_dict={
img_row1: np.expand_dims(np.expand_dims(crush[0], 0), -1),
img_row2: np.expand_dims(np.expand_dims(noncrush[1], 0), -1),
img_row3: np.expand_dims(np.expand_dims(crush[2], 0), -1),
img_row4: np.expand_dims(np.expand_dims(noncrush[3], 0), -1),
img_row5: np.expand_dims(np.expand_dims(crush[4], 0), -1),
img_row6: np.expand_dims(np.expand_dims(noncrush[5], 0), -1),
img_row7: np.expand_dims(np.expand_dims(crush[6], 0), -1),
img_row8: np.expand_dims(np.expand_dims(noncrush[7], 0), -1),
mri_ph: np.expand_dims(np.expand_dims(mri, 0), -1),
label1: np.expand_dims(np.expand_dims(perf[0], 0), -1),
label2: np.expand_dims(np.expand_dims(perf[1], 0), -1),
label3: np.expand_dims(np.expand_dims(perf[2], 0), -1),
label4: np.expand_dims(np.expand_dims(perf[3], 0), -1),
label5: np.expand_dims(np.expand_dims(perf[4], 0), -1),
label6: np.expand_dims(np.expand_dims(perf[5], 0), -1),
label7: np.expand_dims(np.expand_dims(perf[6], 0), -1),
label8: np.expand_dims(np.expand_dims(angio[0], 0), -1),
label9: np.expand_dims(np.expand_dims(angio[1], 0), -1),
label10: np.expand_dims(np.expand_dims(angio[2], 0), -1),
label11: np.expand_dims(np.expand_dims(angio[3], 0), -1),
label12: np.expand_dims(np.expand_dims(angio[4], 0), -1),
label13: np.expand_dims(np.expand_dims(angio[5], 0), -1),
label14: np.expand_dims(np.expand_dims(angio[6], 0), -1),
is_training: False,
input_dim: patch_window,
all_loss: -1.,
})
elapsed = time.time() - t
Elapsed.append(elapsed)
print(elapsed)
for i in range(np.shape(out)[-1]):
image = out[0, :, :, :, i]
sitk_image = sitk.GetImageFromArray(image)
res_dir = test_set[img_indx][0][0].split('/')[-2]
if i == 0:
os.mkdir(parent_path + 'results/' + res_dir)
if i < 7:
nm = 'perf'
else:
nm = 'angi'
sitk_image.SetDirection(direction=direction)
sitk_image.SetOrigin(origin=origin)
sitk_image.SetSpacing(spacing=spacing)
sitk.WriteImage(
sitk_image, parent_path + 'results/' + res_dir + '/' + nm +
'_' + str(i % 7) + '.mha')
print(parent_path + 'results/' + res_dir + '/' + nm + '_' +
str(i % 7) + '.mha done!')
for i in range(7):
if i == 0:
os.mkdir(parent_path + 'results/' + res_dir + '/GT/')
sitk_angio = sitk.GetImageFromArray(angio[i])
sitk_angio.SetDirection(direction=direction)
sitk_angio.SetOrigin(origin=origin)
sitk_angio.SetSpacing(spacing=spacing)
sitk.WriteImage(
sitk_angio, parent_path + 'results/' + res_dir + '/GT/angio_' +
str(i) + '.mha')
sitk_perf = sitk.GetImageFromArray(perf[i])
sitk_perf.SetDirection(direction=direction)
sitk_perf.SetOrigin(origin=origin)
sitk_perf.SetSpacing(spacing=spacing)
sitk.WriteImage(
sitk_perf, parent_path + 'results/' + res_dir + '/GT/perf_' +
str(i) + '.mha')
dic_perf0.append(
anly.analysis(out[0, :, :, :, 0], perf[i], 0, max_perf))
dic_perf1.append(
anly.analysis(out[0, :, :, :, 1], perf[i], 0, max_perf))
dic_perf2.append(
anly.analysis(out[0, :, :, :, 2], perf[i], 0, max_perf))
dic_perf3.append(
anly.analysis(out[0, :, :, :, 3], perf[i], 0, max_perf))
dic_perf4.append(
anly.analysis(out[0, :, :, :, 4], perf[i], 0, max_perf))
dic_perf5.append(
anly.analysis(out[0, :, :, :, 5], perf[i], 0, max_perf))
dic_perf6.append(
anly.analysis(out[0, :, :, :, 6], perf[i], 0, max_perf))
dic_angio0.append(
anly.analysis(out[0, :, :, :, 7], angio[i], 0, max_angio))
dic_angio1.append(
anly.analysis(out[0, :, :, :, 8], angio[i], 0, max_angio))
dic_angio2.append(
anly.analysis(out[0, :, :, :, 9], angio[i], 0, max_angio))
dic_angio3.append(
anly.analysis(out[0, :, :, :, 10], angio[i], 0, max_angio))
dic_angio4.append(
anly.analysis(out[0, :, :, :, 11], angio[i], 0, max_angio))
dic_angio5.append(
anly.analysis(out[0, :, :, :, 12], angio[i], 0, max_angio))
dic_angio6.append(
anly.analysis(out[0, :, :, :, 13], angio[i], 0, max_angio))
if img_indx == 0:
headers = dic_perf0[0].keys()
dics = [
dic_perf0, dic_perf1, dic_perf2, dic_perf3, dic_perf4, dic_perf5,
dic_perf6, dic_angio0, dic_angio1, dic_angio2, dic_angio3,
dic_angio4, dic_angio5, dic_angio6
]
# print(np.mean(Elapsed))
# print(np.std(Elapsed))
save_in_xlsx(parent_path, headers, dics=dics)
print('Total loss: ', loss / len(test_set))
dic_angio3 = []
dic_angio4 = []
dic_angio5 = []
dic_angio6 = []
max_angio = 2.99771428108
max_perf = 17.0151833445
type=["/angi_","/perf_"]
gt_type=["/angio_","/perf_"]
for i in range(2,len(data_dir)):
for j in range(7):
for k in range(2):
output= sitk.GetArrayFromImage(sitk.ReadImage(data_dir[i] + type[k] +str(j)+ ".mha" ))
GT= sitk.GetArrayFromImage(sitk.ReadImage(data_dir[i] +"/GT"+ gt_type[k] +str(j)+ ".mha" ))
if k==1:
if j==0:
dic_perf0.append(anly.analysis(output,GT, 0, max_perf))
elif j==1:
dic_perf1.append(anly.analysis(output, GT, 0, max_perf))
elif j == 2:
dic_perf2.append(anly.analysis(output, GT, 0, max_perf))
elif j==3:
dic_perf3.append(anly.analysis(output, GT, 0, max_perf))
elif j==4:
dic_perf4.append(anly.analysis(output, GT, 0, max_perf))
elif j==5:
dic_perf5.append(anly.analysis(output, GT, 0, max_perf))
elif j==6:
dic_perf6.append(anly.analysis(output, GT, 0, max_perf))
else:
if j==0:
dic_angio0.append(anly.analysis(output,GT, 0, max_angio))
import functions.analysis as anly
import SimpleITK as sitk
if __name__ == '__main__':
path = '/srv/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/sythesize_code/ASL_LOG/multi_stage/'
experiment = 'experiment-2/'
results = 'results/'
subject = '18_PP5_53_cinema2/'
type = 'perf_0.mha'
gt_pth = path + experiment + results + subject + 'GT/' + type
img_pth = path + experiment + results + subject + type
gt = sitk.GetArrayFromImage(sitk.ReadImage(gt_pth))
img = sitk.GetArrayFromImage(sitk.ReadImage(img_pth))
max_angio = 2.99771428108
max_perf = 17.0151833445
print(anly.analysis(img, gt, 0, max_perf))
def test_all_nets(num, test_set):
chckpnt_dir = parent_path + 'unet_checkpoints/'
img_row1 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row2 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row3 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row4 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row5 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row6 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row7 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
img_row8 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
mri_ph = tf.placeholder(tf.float32,
shape=[None, None, None, None, 1],
name='mri')
label1 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label2 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label3 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label4 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label5 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label6 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label7 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label8 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label9 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label10 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label11 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label12 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label13 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
label14 = tf.placeholder(tf.float32, shape=[None, None, None, None, 1])
all_loss = tf.placeholder(tf.float32, name='loss')
is_training = tf.placeholder(tf.bool, name='is_training')
input_dim = tf.placeholder(tf.int32, name='input_dim')
multi_stage_densenet = _multi_stage_densenet()
y, loss_upsampling11, loss_upsampling22 = multi_stage_densenet.multi_stage_densenet(
img_row1=img_row1,
img_row2=img_row2,
img_row3=img_row3,
img_row4=img_row4,
img_row5=img_row5,
img_row6=img_row6,
img_row7=img_row7,
img_row8=img_row8,
input_dim=input_dim,
mri=mri_ph,
is_training=is_training)
loss_instance = _loss_func()
labels = []
labels.append(label1)
labels.append(label2)
labels.append(label3)
labels.append(label4)
labels.append(label5)
labels.append(label6)
labels.append(label7)
labels.append(label8)
labels.append(label9)
labels.append(label10)
labels.append(label11)
labels.append(label12)
labels.append(label13)
labels.append(label14)
logits = []
logits.append(y[:, :, :, :, 0, np.newaxis])
logits.append(y[:, :, :, :, 1, np.newaxis])
logits.append(y[:, :, :, :, 2, np.newaxis])
logits.append(y[:, :, :, :, 3, np.newaxis])
logits.append(y[:, :, :, :, 4, np.newaxis])
logits.append(y[:, :, :, :, 5, np.newaxis])
logits.append(y[:, :, :, :, 6, np.newaxis])
logits.append(y[:, :, :, :, 7, np.newaxis])
logits.append(y[:, :, :, :, 8, np.newaxis])
logits.append(y[:, :, :, :, 9, np.newaxis])
logits.append(y[:, :, :, :, 10, np.newaxis])
logits.append(y[:, :, :, :, 11, np.newaxis])
logits.append(y[:, :, :, :, 12, np.newaxis])
logits.append(y[:, :, :, :, 13, np.newaxis])
stage1 = []
stage1.append(loss_upsampling11[:, :, :, :, 0, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 1, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 2, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 3, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 4, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 5, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 6, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 7, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 8, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 9, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 10, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 11, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 12, np.newaxis])
stage1.append(loss_upsampling11[:, :, :, :, 13, np.newaxis])
stage2 = []
stage2.append(loss_upsampling22[:, :, :, :, 0, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 1, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 2, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 3, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 4, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 5, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 6, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 7, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 8, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 9, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 10, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 11, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 12, np.newaxis])
stage2.append(loss_upsampling22[:, :, :, :, 13, np.newaxis])
with tf.name_scope('Loss'):
loss_dic = loss_instance.loss_selector(
'Multistage_ssim_perf_angio_loss',
labels=labels,
logits=logits,
stage1=stage1,
stage2=stage2)
cost = tf.reduce_mean(loss_dic["loss"], name="cost")
sess = tf.Session()
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(chckpnt_dir)
model_path = ckpt.model_checkpoint_path.rsplit(
'/',
1)[0] + '/unet_inter_epoch0_point' + str(num) + '.ckpt-' + str(num)
saver.restore(sess, model_path)
# copyfile('./test_synthesize_multistage_perf_angio.py', result_path + '/test_synthesize_multistage_perf_angio.py')
_image_class = image_class(train_data,
bunch_of_images_no=1,
is_training=1,
patch_window=patch_window,
sample_no_per_bunch=1,
label_patch_size=label_patchs_size,
validation_total_sample=0)
learning_rate = 1E-5
extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(extra_update_ops):
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost)
# init = tf.global_variables_initializer()
loss = 0
for img_indx in range(len(test_set[0:1])):
crush, noncrush, perf, angio, mri, segmentation_, spacing, direction, origin = _image_class.read_image_for_test(
test_set=test_set,
img_indx=img_indx,
input_size=in_dim,
final_layer=final_layer)
[out] = sess.run(
[y],
feed_dict={
img_row1: np.expand_dims(np.expand_dims(crush[0], 0), -1),
img_row2: np.expand_dims(np.expand_dims(noncrush[1], 0), -1),
img_row3: np.expand_dims(np.expand_dims(crush[2], 0), -1),
img_row4: np.expand_dims(np.expand_dims(noncrush[3], 0), -1),
img_row5: np.expand_dims(np.expand_dims(crush[4], 0), -1),
img_row6: np.expand_dims(np.expand_dims(noncrush[5], 0), -1),
img_row7: np.expand_dims(np.expand_dims(crush[6], 0), -1),
img_row8: np.expand_dims(np.expand_dims(noncrush[7], 0), -1),
mri_ph: np.expand_dims(np.expand_dims(mri, 0), -1),
label1: np.expand_dims(np.expand_dims(perf[0], 0), -1),
label2: np.expand_dims(np.expand_dims(perf[1], 0), -1),
label3: np.expand_dims(np.expand_dims(perf[2], 0), -1),
label4: np.expand_dims(np.expand_dims(perf[3], 0), -1),
label5: np.expand_dims(np.expand_dims(perf[4], 0), -1),
label6: np.expand_dims(np.expand_dims(perf[5], 0), -1),
label7: np.expand_dims(np.expand_dims(perf[6], 0), -1),
label8: np.expand_dims(np.expand_dims(angio[0], 0), -1),
label9: np.expand_dims(np.expand_dims(angio[1], 0), -1),
label10: np.expand_dims(np.expand_dims(angio[2], 0), -1),
label11: np.expand_dims(np.expand_dims(angio[3], 0), -1),
label12: np.expand_dims(np.expand_dims(angio[4], 0), -1),
label13: np.expand_dims(np.expand_dims(angio[5], 0), -1),
label14: np.expand_dims(np.expand_dims(angio[6], 0), -1),
is_training: False,
input_dim: patch_window,
all_loss: -1.,
})
for i in range(np.shape(out)[-1]):
image = out[0, :, :, :, i]
gt = sitk.GetArrayFromImage(sitk.GetImageFromArray(perf[i]))
dic = analysis.analysis(result=image, gt=gt, min=0, max=max_perf)
print(dic)
# for i in range(np.shape(out)[-1]):
# image = out[0, :, :, :, i]
# sitk_image = sitk.GetImageFromArray(image)
# res_dir = test_set[img_indx][0][0].split('/')[-2]
# if i == 0:
# os.mkdir(parent_path + 'results/' + res_dir)
# if i < 7:
# nm = 'perf'
# else:
# nm = 'angi'
# sitk_image.SetDirection(direction=direction)
# sitk_image.SetOrigin(origin=origin)
# sitk_image.SetSpacing(spacing=spacing)
# sitk.WriteImage(sitk_image, parent_path + 'results/' + res_dir + '/' + nm + '_' + str(i % 7) + '.mha')
# print(parent_path + 'results/' + res_dir + '/' + nm + '_' + str(i % 7) + '.mha done!')
# for i in range(7):
# if i == 0:
# os.mkdir(parent_path + 'results/' + res_dir + '/GT/')
# sitk_angio = sitk.GetImageFromArray(angio[i])
# sitk_angio.SetDirection(direction=direction)
# sitk_angio.SetOrigin(origin=origin)
# sitk_angio.SetSpacing(spacing=spacing)
# sitk.WriteImage(sitk_angio, parent_path + 'results/' + res_dir + '/GT/angio_' + str(i) + '.mha')
#
# sitk_perf = sitk.GetImageFromArray(perf[i])
# sitk_perf.SetDirection(direction=direction)
# sitk_perf.SetOrigin(origin=origin)
# sitk_perf.SetSpacing(spacing=spacing)
# sitk.WriteImage(sitk_perf, parent_path + 'results/' + res_dir + '/GT/perf_' + str(i) + '.mha')
a = 1
# plt.imshow(out[0, int(gt_cube_size / 2), :, :, 0])
# plt.figure()
# loss += loss_train1
# print('Loss_train: ', loss_train1)
print('Total loss: ', loss / len(test_set))
def test_all_nets():
data = 2
Server = 'DL'
if Server == 'DL':
parent_path = '/srv/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/sythesize_code/ASL_LOG/Log_perceptual/regularization/perceptual-0/'
data_path = '/srv/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/Data-01/BrainWeb_permutation00_low/'
else:
parent_path = '/exports/lkeb-hpc/syousefi/Code/'
data_path = '/srv/2-lkeb-17-dl01/syousefi/TestCode/EsophagusProject/Data-01/BrainWeb_permutation00_low/'
img_name = ''
label_name = ''
_rd = _read_data(data=data,
reverse=False,
img_name=img_name,
label_name=label_name,
dataset_path=data_path)
'''read path of the images for train, test, and validation'''
train_data, validation_data, test_data = _rd.read_data_path()
chckpnt_dir = parent_path + 'unet_checkpoints/'
result_path = parent_path + 'results/'
batch_no = 1
batch_no_validation = batch_no
# label_patchs_size = 87#39 # 63
# patch_window = 103#53 # 77#89
if test_vali == 1:
test_set = validation_data
else:
test_set = test_data
# ===================================================================================
img_row1 = tf.placeholder(
tf.float32,
shape=[batch_no, patch_window, patch_window, patch_window, 1],
name='img_row1')
img_row2 = tf.placeholder(
tf.float32,
shape=[batch_no, patch_window, patch_window, patch_window, 1],
name='img_row2')
img_row3 = tf.placeholder(
tf.float32,
shape=[batch_no, patch_window, patch_window, patch_window, 1],
name='img_row3')
img_row4 = tf.placeholder(
tf.float32,
shape=[batch_no, patch_window, patch_window, patch_window, 1],
name='img_row4')
img_row5 = tf.placeholder(
tf.float32,
shape=[batch_no, patch_window, patch_window, patch_window, 1],
name='img_row5')
img_row6 = tf.placeholder(
tf.float32,
shape=[batch_no, patch_window, patch_window, patch_window, 1],
name='img_row6')
img_row7 = tf.placeholder(
tf.float32,
shape=[batch_no, patch_window, patch_window, patch_window, 1],
name='img_row7')
img_row8 = tf.placeholder(
tf.float32,
shape=[batch_no, patch_window, patch_window, patch_window, 1],
name='img_row8')
mri_ph = tf.placeholder(
tf.float32,
shape=[batch_no, patch_window, patch_window, patch_window, 1],
name='mri')
label1 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size, label_patchs_size,
label_patchs_size, 1
],
name='label1')
label2 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size, label_patchs_size,
label_patchs_size, 1
],
name='label2')
label3 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size, label_patchs_size,
label_patchs_size, 1
],
name='label3')
label4 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size, label_patchs_size,
label_patchs_size, 1
],
name='label4')
label5 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size, label_patchs_size,
label_patchs_size, 1
],
name='label5')
label6 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size, label_patchs_size,
label_patchs_size, 1
],
name='label6')
label7 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size, label_patchs_size,
label_patchs_size, 1
],
name='label7')
label8 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size, label_patchs_size,
label_patchs_size, 1
],
name='label8')
label9 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size, label_patchs_size,
label_patchs_size, 1
],
name='label9')
label10 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size,
label_patchs_size, label_patchs_size, 1
],
name='label10')
label11 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size,
label_patchs_size, label_patchs_size, 1
],
name='label11')
label12 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size,
label_patchs_size, label_patchs_size, 1
],
name='label12')
label13 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size,
label_patchs_size, label_patchs_size, 1
],
name='label13')
label14 = tf.placeholder(tf.float32,
shape=[
batch_no, label_patchs_size,
label_patchs_size, label_patchs_size, 1
],
name='label14')
is_training = tf.placeholder(tf.bool, name='is_training')
input_dim = tf.placeholder(tf.int32, name='input_dim')
perf_vgg_loss_tens = tf.placeholder(tf.float32, name='VGG_perf')
angio_vgg_loss_tens = tf.placeholder(tf.float32, name='VGG_angio')
perf_vgg_tens0 = tf.placeholder(tf.float32, name='vgg_perf0')
perf_vgg_tens1 = tf.placeholder(tf.float32, name='vgg_perf1')
perf_vgg_tens2 = tf.placeholder(tf.float32, name='vgg_perf2')
perf_vgg_tens3 = tf.placeholder(tf.float32, name='vgg_perf3')
perf_vgg_tens4 = tf.placeholder(tf.float32, name='vgg_perf4')
perf_vgg_tens5 = tf.placeholder(tf.float32, name='vgg_perf5')
perf_vgg_tens6 = tf.placeholder(tf.float32, name='vgg_perf6')
angio_vgg_tens0 = tf.placeholder(tf.float32, name='vgg_angio0')
angio_vgg_tens1 = tf.placeholder(tf.float32, name='vgg_angio1')
angio_vgg_tens2 = tf.placeholder(tf.float32, name='vgg_angio2')
angio_vgg_tens3 = tf.placeholder(tf.float32, name='vgg_angio3')
angio_vgg_tens4 = tf.placeholder(tf.float32, name='vgg_angio4')
angio_vgg_tens5 = tf.placeholder(tf.float32, name='vgg_angio5')
angio_vgg_tens6 = tf.placeholder(tf.float32, name='vgg_angio6')
perf_huber_loss_tens = tf.placeholder(tf.float32, name='huber_perf')
angio_huber_loss_tens = tf.placeholder(tf.float32, name='huber_angio')
perf_huber_tens0 = tf.placeholder(tf.float32, name='huber_perf0')
perf_huber_tens1 = tf.placeholder(tf.float32, name='huber_perf1')
perf_huber_tens2 = tf.placeholder(tf.float32, name='huber_perf2')
perf_huber_tens3 = tf.placeholder(tf.float32, name='huber_perf3')
perf_huber_tens4 = tf.placeholder(tf.float32, name='huber_perf4')
perf_huber_tens5 = tf.placeholder(tf.float32, name='huber_perf5')
perf_huber_tens6 = tf.placeholder(tf.float32, name='huber_perf6')
angio_huber_tens0 = tf.placeholder(tf.float32, name='huber_angio0')
angio_huber_tens1 = tf.placeholder(tf.float32, name='huber_angio1')
angio_huber_tens2 = tf.placeholder(tf.float32, name='huber_angio2')
angio_huber_tens3 = tf.placeholder(tf.float32, name='huber_angio3')
angio_huber_tens4 = tf.placeholder(tf.float32, name='huber_angio4')
angio_huber_tens5 = tf.placeholder(tf.float32, name='huber_angio5')
angio_huber_tens6 = tf.placeholder(tf.float32, name='huber_angio6')
# ===================================================================================
densenet = _densenet()
[y, _] = densenet.densenet(img_row1=img_row1,
img_row2=img_row2,
img_row3=img_row3,
img_row4=img_row4,
img_row5=img_row5,
img_row6=img_row6,
img_row7=img_row7,
img_row8=img_row8,
input_dim=input_dim,
is_training=is_training)
vgg = vgg_feature_maker(test=1)
feature_type = 'huber'
vgg_y0 = vgg.feed_img(y[:, :, :, :, 0], feature_type=feature_type).copy()
vgg_y1 = vgg.feed_img(y[:, :, :, :, 1], feature_type=feature_type).copy()
vgg_y2 = vgg.feed_img(y[:, :, :, :, 2], feature_type=feature_type).copy()
vgg_y3 = vgg.feed_img(y[:, :, :, :, 3], feature_type=feature_type).copy()
vgg_y4 = vgg.feed_img(y[:, :, :, :, 4], feature_type=feature_type).copy()
vgg_y5 = vgg.feed_img(y[:, :, :, :, 5], feature_type=feature_type).copy()
vgg_y6 = vgg.feed_img(y[:, :, :, :, 6], feature_type=feature_type).copy()
vgg_y7 = vgg.feed_img(y[:, :, :, :, 7], feature_type=feature_type)
vgg_y8 = vgg.feed_img(y[:, :, :, :, 8], feature_type=feature_type)
vgg_y9 = vgg.feed_img(y[:, :, :, :, 9], feature_type=feature_type)
vgg_y10 = vgg.feed_img(y[:, :, :, :, 10], feature_type=feature_type)
vgg_y11 = vgg.feed_img(y[:, :, :, :, 11], feature_type=feature_type)
vgg_y12 = vgg.feed_img(y[:, :, :, :, 12], feature_type=feature_type)
vgg_y13 = vgg.feed_img(y[:, :, :, :, 13], feature_type=feature_type)
vgg_label0 = vgg.feed_img(label1[:, :, :, :, 0],
feature_type=feature_type).copy()
vgg_label1 = vgg.feed_img(label2[:, :, :, :, 0],
feature_type=feature_type).copy()
vgg_label2 = vgg.feed_img(label3[:, :, :, :, 0],
feature_type=feature_type).copy()
vgg_label3 = vgg.feed_img(label4[:, :, :, :, 0],
feature_type=feature_type).copy()
vgg_label4 = vgg.feed_img(label5[:, :, :, :, 0],
feature_type=feature_type).copy()
vgg_label5 = vgg.feed_img(label6[:, :, :, :, 0],
feature_type=feature_type).copy()
vgg_label6 = vgg.feed_img(label7[:, :, :, :, 0],
feature_type=feature_type).copy()
vgg_label7 = vgg.feed_img(label8[:, :, :, :, 0], feature_type=feature_type)
vgg_label8 = vgg.feed_img(label9[:, :, :, :, 0], feature_type=feature_type)
vgg_label9 = vgg.feed_img(label10[:, :, :, :, 0],
feature_type=feature_type)
vgg_label10 = vgg.feed_img(label11[:, :, :, :, 0],
feature_type=feature_type)
vgg_label11 = vgg.feed_img(label12[:, :, :, :, 0],
feature_type=feature_type)
vgg_label12 = vgg.feed_img(label13[:, :, :, :, 0],
feature_type=feature_type)
vgg_label13 = vgg.feed_img(label14[:, :, :, :, 0],
feature_type=feature_type)
all_loss = tf.placeholder(tf.float32, name='loss')
# is_training = tf.placeholder(tf.bool, name='is_training')
# input_dim = tf.placeholder(tf.int32, name='input_dim')
# ave_huber = tf.placeholder(tf.float32, name='huber')
labels = []
labels.append(label1)
labels.append(label2)
labels.append(label3)
labels.append(label4)
labels.append(label5)
labels.append(label6)
labels.append(label7)
labels.append(label8)
labels.append(label9)
labels.append(label10)
labels.append(label11)
labels.append(label12)
labels.append(label13)
labels.append(label14)
logits = []
logits.append(y[:, :, :, :, 0, np.newaxis])
logits.append(y[:, :, :, :, 1, np.newaxis])
logits.append(y[:, :, :, :, 2, np.newaxis])
logits.append(y[:, :, :, :, 3, np.newaxis])
logits.append(y[:, :, :, :, 4, np.newaxis])
logits.append(y[:, :, :, :, 5, np.newaxis])
logits.append(y[:, :, :, :, 6, np.newaxis])
logits.append(y[:, :, :, :, 7, np.newaxis])
logits.append(y[:, :, :, :, 8, np.newaxis])
logits.append(y[:, :, :, :, 9, np.newaxis])
logits.append(y[:, :, :, :, 10, np.newaxis])
logits.append(y[:, :, :, :, 11, np.newaxis])
logits.append(y[:, :, :, :, 12, np.newaxis])
logits.append(y[:, :, :, :, 13, np.newaxis])
loss_instance = _loss_func()
vgg_in_feature = []
vgg_in_feature.append(vgg_y0)
vgg_in_feature.append(vgg_y1)
vgg_in_feature.append(vgg_y2)
vgg_in_feature.append(vgg_y3)
vgg_in_feature.append(vgg_y4)
vgg_in_feature.append(vgg_y5)
vgg_in_feature.append(vgg_y6)
vgg_in_feature.append(vgg_y7)
vgg_in_feature.append(vgg_y8)
vgg_in_feature.append(vgg_y9)
vgg_in_feature.append(vgg_y10)
vgg_in_feature.append(vgg_y11)
vgg_in_feature.append(vgg_y12)
vgg_in_feature.append(vgg_y13)
vgg_label_feature = []
vgg_label_feature.append(vgg_label0)
vgg_label_feature.append(vgg_label1)
vgg_label_feature.append(vgg_label2)
vgg_label_feature.append(vgg_label3)
vgg_label_feature.append(vgg_label4)
vgg_label_feature.append(vgg_label5)
vgg_label_feature.append(vgg_label6)
vgg_label_feature.append(vgg_label7)
vgg_label_feature.append(vgg_label8)
vgg_label_feature.append(vgg_label9)
vgg_label_feature.append(vgg_label10)
vgg_label_feature.append(vgg_label11)
vgg_label_feature.append(vgg_label12)
vgg_label_feature.append(vgg_label13)
with tf.name_scope('Loss'):
loss_dic = loss_instance.loss_selector(
'content_vgg_pairwise_loss_huber',
labels=vgg_label_feature,
logits=vgg_in_feature,
vgg=vgg,
h_labels=labels,
h_logits=logits)
cost = tf.reduce_mean(loss_dic["loss"], name="cost")
# cost_angio = tf.reduce_mean(loss_dic["angio_SSIM"], name="angio_SSIM")
# cost_perf = tf.reduce_mean(loss_dic["perf_SSIM"], name="perf_SSIM")
# ========================================================================
# ave_loss = tf.placeholder(tf.float32, name='loss')
# ave_loss_perf = tf.placeholder(tf.float32, name='loss_perf')
# ave_loss_angio = tf.placeholder(tf.float32, name='loss_angio')
#
# average_gradient_perf = tf.placeholder(tf.float32, name='grad_ave_perf')
# average_gradient_angio = tf.placeholder(tf.float32, name='grad_ave_angio')
#
# ave_huber = tf.placeholder(tf.float32, name='huber')
# restore the model
sess = tf.Session()
saver = tf.train.Saver()
ckpt = tf.train.get_checkpoint_state(chckpnt_dir)
saver.restore(sess, ckpt.model_checkpoint_path)
copyfile('./test_synthesize_ssim_perf_angio.py',
result_path + '/test_synthesize_ssim_perf_angio.py')
_image_class = image_class(train_data,
bunch_of_images_no=1,
is_training=1,
patch_window=patch_window,
sample_no_per_bunch=1,
label_patch_size=label_patchs_size,
validation_total_sample=0)
learning_rate = 1E-5
extra_update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
with tf.control_dependencies(extra_update_ops):
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost)
# init = tf.global_variables_initializer()
dic_perf0 = []
dic_perf1 = []
dic_perf2 = []
dic_perf3 = []
dic_perf4 = []
dic_perf5 = []
dic_perf6 = []
dic_angio0 = []
dic_angio1 = []
dic_angio2 = []
dic_angio3 = []
dic_angio4 = []
dic_angio5 = []
dic_angio6 = []
loss = 0
for img_indx in range(len(test_set)):
crush, noncrush, perf, angio, mri, segmentation_, spacing, direction, origin = _image_class.read_image_for_test(
test_set=test_set,
img_indx=img_indx,
input_size=in_dim,
final_layer=final_layer)
[out] = sess.run(
[y],
feed_dict={
img_row1: np.expand_dims(np.expand_dims(crush[0], 0), -1),
img_row2: np.expand_dims(np.expand_dims(noncrush[1], 0), -1),
img_row3: np.expand_dims(np.expand_dims(crush[2], 0), -1),
img_row4: np.expand_dims(np.expand_dims(noncrush[3], 0), -1),
img_row5: np.expand_dims(np.expand_dims(crush[4], 0), -1),
img_row6: np.expand_dims(np.expand_dims(noncrush[5], 0), -1),
img_row7: np.expand_dims(np.expand_dims(crush[6], 0), -1),
img_row8: np.expand_dims(np.expand_dims(noncrush[7], 0), -1),
mri_ph: np.expand_dims(np.expand_dims(mri, 0), -1),
label1: np.expand_dims(np.expand_dims(perf[0], 0), -1),
label2: np.expand_dims(np.expand_dims(perf[1], 0), -1),
label3: np.expand_dims(np.expand_dims(perf[2], 0), -1),
label4: np.expand_dims(np.expand_dims(perf[3], 0), -1),
label5: np.expand_dims(np.expand_dims(perf[4], 0), -1),
label6: np.expand_dims(np.expand_dims(perf[5], 0), -1),
label7: np.expand_dims(np.expand_dims(perf[6], 0), -1),
label8: np.expand_dims(np.expand_dims(angio[0], 0), -1),
label9: np.expand_dims(np.expand_dims(angio[1], 0), -1),
label10: np.expand_dims(np.expand_dims(angio[2], 0), -1),
label11: np.expand_dims(np.expand_dims(angio[3], 0), -1),
label12: np.expand_dims(np.expand_dims(angio[4], 0), -1),
label13: np.expand_dims(np.expand_dims(angio[5], 0), -1),
label14: np.expand_dims(np.expand_dims(angio[6], 0), -1),
is_training: False,
input_dim: patch_window,
all_loss: -1.,
angio_vgg_loss_tens: -1, # vgg angio
perf_vgg_loss_tens: -1,
perf_vgg_tens0: -1,
perf_vgg_tens1: -1,
perf_vgg_tens2: -1,
perf_vgg_tens3: -1,
perf_vgg_tens4: -1,
perf_vgg_tens5: -1,
perf_vgg_tens6: -1,
angio_vgg_tens0: -1,
angio_vgg_tens1: -1,
angio_vgg_tens2: -1,
angio_vgg_tens3: -1,
angio_vgg_tens4: -1,
angio_vgg_tens5: -1,
angio_vgg_tens6: -1,
perf_huber_loss_tens: -1,
angio_huber_loss_tens: -1,
perf_huber_tens0: -1,
perf_huber_tens1: -1,
perf_huber_tens2: -1,
perf_huber_tens3: -1,
perf_huber_tens4: -1,
perf_huber_tens5: -1,
perf_huber_tens6: -1,
angio_huber_tens0: -1,
angio_huber_tens1: -1,
angio_huber_tens2: -1,
angio_huber_tens3: -1,
angio_huber_tens4: -1,
angio_huber_tens5: -1,
angio_huber_tens6: -1,
})
for i in range(np.shape(out)[-1]):
image = out[0, :, :, :, i]
sitk_image = sitk.GetImageFromArray(image)
res_dir = test_set[img_indx][0][0].split('/')[-2]
if i == 0:
os.mkdir(parent_path + 'results/' + res_dir)
if i < 7:
nm = 'perf'
else:
nm = 'angi'
sitk_image.SetDirection(direction=direction)
sitk_image.SetOrigin(origin=origin)
sitk_image.SetSpacing(spacing=spacing)
sitk.WriteImage(
sitk_image, parent_path + 'results/' + res_dir + '/' + nm +
'_' + str(i % 7) + '.mha')
print(parent_path + 'results/' + res_dir + '/' + nm + '_' +
str(i % 7) + '.mha done!')
for i in range(7):
if i == 0:
os.mkdir(parent_path + 'results/' + res_dir + '/GT/')
sitk_angio = sitk.GetImageFromArray(angio[i])
sitk_angio.SetDirection(direction=direction)
sitk_angio.SetOrigin(origin=origin)
sitk_angio.SetSpacing(spacing=spacing)
sitk.WriteImage(
sitk_angio, parent_path + 'results/' + res_dir + '/GT/angio_' +
str(i) + '.mha')
sitk_perf = sitk.GetImageFromArray(perf[i])
sitk_perf.SetDirection(direction=direction)
sitk_perf.SetOrigin(origin=origin)
sitk_perf.SetSpacing(spacing=spacing)
sitk.WriteImage(
sitk_perf, parent_path + 'results/' + res_dir + '/GT/perf_' +
str(i) + '.mha')
a = 1
dic_perf0.append(
anly.analysis(out[0, :, :, :, 0], perf[i], 0, max_perf))
dic_perf1.append(
anly.analysis(out[0, :, :, :, 1], perf[i], 0, max_perf))
dic_perf2.append(
anly.analysis(out[0, :, :, :, 2], perf[i], 0, max_perf))
dic_perf3.append(
anly.analysis(out[0, :, :, :, 3], perf[i], 0, max_perf))
dic_perf4.append(
anly.analysis(out[0, :, :, :, 4], perf[i], 0, max_perf))
dic_perf5.append(
anly.analysis(out[0, :, :, :, 5], perf[i], 0, max_perf))
dic_perf6.append(
anly.analysis(out[0, :, :, :, 6], perf[i], 0, max_perf))
dic_angio0.append(
anly.analysis(out[0, :, :, :, 7], angio[i], 0, max_angio))
dic_angio1.append(
anly.analysis(out[0, :, :, :, 8], angio[i], 0, max_angio))
dic_angio2.append(
anly.analysis(out[0, :, :, :, 9], angio[i], 0, max_angio))
dic_angio3.append(
anly.analysis(out[0, :, :, :, 10], angio[i], 0, max_angio))
dic_angio4.append(
anly.analysis(out[0, :, :, :, 11], angio[i], 0, max_angio))
dic_angio5.append(
anly.analysis(out[0, :, :, :, 12], angio[i], 0, max_angio))
dic_angio6.append(
anly.analysis(out[0, :, :, :, 13], angio[i], 0, max_angio))
if img_indx == 0:
headers = dic_perf0[0].keys()
dics = [
dic_perf0, dic_perf1, dic_perf2, dic_perf3, dic_perf4, dic_perf5,
dic_perf6, dic_angio0, dic_angio1, dic_angio2, dic_angio3,
dic_angio4, dic_angio5, dic_angio6
]
save_in_xlsx(parent_path, headers, dics=dics)
# plt.imshow(out[0, int(gt_cube_size / 2), :, :, 0])
# plt.figure()
# loss += loss_train1
# print('Loss_train: ', loss_train1)
print('Total loss: ', loss / len(test_set))