class TransferFbUpTrain(object):
def __init__(self,
sess,
input_placer,
gold_placer,
learn_rate_placer,
keep_prob,
window_size=15,
model_path='./model/version_15',
image_path='./data/neuron/train',
valid_flag=False,
valid_num=2048,
nn=FbUp(),
predict_flag=True,
wd=0.001,
wdo=0.001):
self.sess = sess
self.batch_generator = None
self.nn = nn
self.saver = None
self.predict_flag = predict_flag
self.fg_para_dict = dict()
self.valid_flag = valid_flag
self.window_size = window_size
if valid_flag:
self.batch_generator = ImageListBatchGenerator(
(window_size - 1) // 2, image_path=image_path)
self.valid = self.next_batch(valid_num)
self.get_nn(input_placer,
gold_placer,
learn_rate_placer,
keep_prob,
wd=wd,
wdo=wdo,
window_size=window_size)
print("Creation Complete\nIntialization")
init_op = tf.initialize_all_variables()
print("Start Intialization")
sess.run(init_op)
print("Intialization Done")
self.restore(model_path)
print("Saving")
#self.saver = tf.train.Saver()
#self.saver.save(self.sess,model_path + '/transfer_model.ckpt')
#print("Saved at " + model_path + '/transfer_model.ckpt')
def get_nn(self,
input_placer,
gold_placer,
learn_rate_placer,
keep_prob,
window_size,
conv_1_neuron_num=256,
conv_2_neuron_num=128,
conv_3_neuron_num=64,
conv_4_neuron_num=64,
conv_5_neuron_num=64,
conv_1=1,
conv_2=3,
conv_3=3,
conv_4=3,
conv_5=1,
wd=1e-9,
wdo=1e-9):
#create sub classes
self.x = input_placer
self.y_ = gold_placer
self.l_rate = learn_rate_placer
self.keep_prob = keep_prob
fg_dict = dict()
y_conv_combined = []
y_res_combined = []
for angle_xy in range(30, 181, 30):
for angle_xz in range(30, 181, 30):
print("Create FG NN %d %d" % (angle_xy, angle_xz))
fg_dict[(angle_xy,angle_xz)] = \
FbUpTrain(self.sess,input_placer,gold_placer,\
keep_prob,learn_rate_placer,\
angle_xy = angle_xy, \
angle_xz = angle_xz,
train_flag = False,
wd = 0e-9)
fg_variables = tf.get_collection(
tf.GraphKeys.TRAINABLE_VARIABLES,
scope="fg_" + str(angle_xy) + '_' + str(angle_xz))
fg_dict[(angle_xy,angle_xz)].saver = \
tf.train.Saver(fg_variables)
#fg_dict[(angle_xy,angle_xz)].saver.restore(self.sess,\
# model_path + '/model_foreground_'+str(angle_xy)+'_'+str(angle_xz)+'.ckpt')
y_conv_combined.append(
tf.expand_dims(
fg_dict[(angle_xy, angle_xz)].para_dict['h_conv4'], 5))
#y_res_combined.append(
# tf.expand_dims(fg_dict[(angle_xy,angle_xz)].y_res,4))
FEAT_NUM = 32
self.fg_dict = fg_dict
y_conv_combined = tf.concat(5, y_conv_combined)
shape = tf.shape(y_conv_combined)
dir_out = tf.reshape(
y_conv_combined,
[shape[0], shape[1], shape[2], shape[3], 36 * FEAT_NUM])
#dir_out = tf.reshape(y_conv_combined,[-1,window_size,window_size,window_size,36 * FEAT_NUM])
self.dir_out = dir_out
with tf.variable_scope("transfer"):
#h_pool2 = self.pool_3x3(h_conv2)
with tf.variable_scope("transfer_layer_1"):
W_conv1 = self.nn.weight_variable(
[conv_1, conv_1, conv_1, 36 * FEAT_NUM, conv_1_neuron_num],
wd, 1**3 * 36 * FEAT_NUM, conv_1**3 * conv_1_neuron_num)
b_conv1 = self.nn.bias_variable([conv_1_neuron_num])
h_conv1 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(dir_out, W_conv1), b_conv1))
with tf.variable_scope("transfer_layer_2"):
W_conv2 = self.nn.weight_variable([
conv_2, conv_2, conv_2, conv_1_neuron_num,
conv_2_neuron_num
], wd, conv_1**3 * conv_1_neuron_num, conv_2**3 *
conv_2_neuron_num)
b_conv2 = self.nn.bias_variable([conv_2_neuron_num])
h_conv2 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv1, W_conv2), b_conv2))
with tf.variable_scope("transfer_layer_3"):
W_conv3 = self.nn.weight_variable([
conv_3, conv_3, conv_3, conv_2_neuron_num,
conv_3_neuron_num
], wd, conv_2**3 * conv_2_neuron_num, conv_3**3 *
conv_3_neuron_num)
b_conv3 = self.nn.bias_variable([conv_3_neuron_num])
h_conv3 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv2, W_conv3), b_conv3))
with tf.variable_scope("transfer_layer_4"):
W_conv4 = self.nn.weight_variable([
conv_4, conv_4, conv_4, conv_3_neuron_num,
conv_4_neuron_num
], wd, conv_3**3 * conv_3_neuron_num, conv_4**3 *
conv_4_neuron_num)
b_conv4 = self.nn.bias_variable([conv_4_neuron_num])
h_conv4 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv3, W_conv4), b_conv4))
with tf.variable_scope("transfer_layer_5"):
W_conv5 = self.nn.weight_variable([
conv_5, conv_5, conv_5, conv_4_neuron_num,
conv_5_neuron_num
], wd, conv_4**3 * conv_4_neuron_num, conv_5**3 *
conv_5_neuron_num)
b_conv5 = self.nn.bias_variable([conv_5_neuron_num])
h_conv5 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv4, W_conv5), b_conv5))
weight_decay = tf.mul(tf.nn.l2_loss(h_conv5),
wd,
name='weight_loss')
tf.add_to_collection('losses', weight_decay)
with tf.variable_scope("transfer_final"):
W_fc1 = self.nn.weight_variable(
[1, 1, 1, conv_5_neuron_num, 2], wd=wdo)
b_fc1 = self.nn.bias_variable([2])
h_fc1 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv5, W_fc1), b_fc1))
if self.predict_flag:
final_shape = tf.shape(h_fc1)
self.y_conv = tf.nn.softmax(h_fc1[:, (final_shape[1] - 1) / 2,
(final_shape[2] - 1) / 2,
(final_shape[3] - 1) / 2, :])
#self.y_conv = tf.nn.softmax(h_fc1[:,(window_size-1)//2,(window_size-1)//2,(window_size-1)//2,:])
self.y_res = tf.argmax(self.y_conv, 1)
else:
curr_shape = tf.shape(h_fc1)
reshaped_conv = tf.reshape(h_fc1, [-1, 2])
softmaxed_reshaped_conv = tf.nn.softmax(reshaped_conv)
self.y_conv = tf.reshape(softmaxed_reshaped_conv, curr_shape)
self.y_res = tf.argmax(self.y_conv, 4)
self.y_conv_2x = dilation3D(self.y_conv)
tmp = tf.expand_dims(self.y_res, 4)
self.y_res_2x = tf.squeeze(dilation3D(tmp))
if self.valid_flag:
cross_entropy_mean = -tf.reduce_mean(self.y_ * tf.log(self.y_conv))
tf.add_to_collection('losses', cross_entropy_mean)
cross_entropy = tf.add_n(tf.get_collection('losses'),
name='total_loss')
transfer_vars = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,
"transfer")
self.train_step = tf.train.AdamOptimizer(self.l_rate).minimize(
cross_entropy, var_list=transfer_vars)
correct_prediction = tf.equal(tf.argmax(self.y_conv, 1),
tf.argmax(self.y_, 1))
self.accuracy = tf.reduce_mean(tf.cast(correct_prediction,
"float"))
else:
print("Transfer Not For Training Singly")
def restore(self, model_path):
print("Restoring...")
for angle_xy in range(30, 181, 30):
for angle_xz in range(30, 181, 30):
print("Restore %d %d" % (angle_xy, angle_xz))
self.fg_dict[(angle_xy,angle_xz)].saver.restore(self.sess,\
model_path + '/model_foreground_'+str(angle_xy)+'_'+str(angle_xz)+'.ckpt')
print("Done")
def next_batch(self, batch_size):
return self.batch_generator.next_batch(batch_size)
def get_valid_set(self):
return self.valid
def train(self,
epic_num=8,
loop_size=2100,
batch_size=32,
learning_rate=0.01,
thres=0.01,
keep_prob=1.0,
tao=50):
accuracies = [0.0, 0.05, 0.1, 0.15, 0.2, .25, .3, .35, .4, .45, .5]
tao_counter = 0
for epic in range(epic_num):
for i in range(loop_size):
#ts = time.time()
batch = self.next_batch(batch_size)
#te = time.time()
#print('Gen Batch: ', te-ts)
if i % 10 == 0:
print(learning_rate)
train_accuracy = 0
valid = self.get_valid_set()
print(np.max(valid[0]))
valid_counter = 0
for v_idx in range(0, len(valid[0]), 256):
train_accuracy += self.accuracy.eval(
session=self.sess,
feed_dict={
self.x:
valid[0][v_idx:min(v_idx + 256, len(valid[0]))]
/ 255.0,
self.y_:
valid[1][v_idx:min(v_idx +
256, len(valid[0]))],
self.keep_prob:
1.0,
self.l_rate:
learning_rate
})
valid_counter += 1
dir_out = self.dir_out.eval(
session=self.sess,
feed_dict={
self.x:
valid[0][v_idx:min(v_idx + 256, len(valid[0]))]
/ 255.0,
self.y_:
valid[1][v_idx:min(v_idx +
256, len(valid[0]))],
self.keep_prob:
1.0,
self.l_rate:
learning_rate
})
print(dir_out.shape, np.max(dir_out), np.min(dir_out))
train_accuracy /= valid_counter
tao_counter += 1
#print corrects
print("%d step %d, training accuracy %g" %
(epic, i, train_accuracy))
if np.std(accuracies[-10:]) < thres or tao_counter > tao:
learning_rate = max(1e-6, learning_rate / 2)
thres /= 2
accuracies = [
0.0, 0.05, 0.1, 0.15, 0.2, .25, .3, .35, .4, .45,
.5
]
tao_counter = 0
accuracies.append(train_accuracy)
#corrects = correct_prediction.eval(feed_dict={x1: batch[0],x2: batch[1], y_: batch[2], keep_prob: 1.0,l_rate: learn_rate * (4.0-epic)})
#reinforce = append_reinforce(corrects,batch,reinforce)
#ts = time.time()
self.train_step.run(session=self.sess,
feed_dict={
self.x: batch[0] / 255.0,
self.y_: batch[1],
self.keep_prob: keep_prob,
self.l_rate: learning_rate
})
#te = time.time()
#print('Train: ',te-ts)
def predict_large(self,
image,
version='',
batch_diameter=64,
extra_edge_width=15):
sx, sy, sz = image.shape
print(sx, sy, sz)
#batches,pos,neg = sampling(image)
result = np.zeros((sx, sy, sz))
max_res = np.zeros((sx, sy, sz))
for x_start, y_start, z_start, batch in image3D_crop_generator(
image, batch_diameter, extra_edge_width):
ts = time.time()
num, x_len, y_len, z_len, _ = batch.shape
#print(batch.shape)
tmp_res = self.y_res_2x.eval(session=self.sess,
feed_dict={
self.x: batch / 255.0,
self.y_: np.zeros((num, 2)),
self.keep_prob: 1.0,
self.l_rate: 0
})
print(tmp_res.shape)
tmp_res = tmp_res[extra_edge_width:-extra_edge_width,
extra_edge_width:-extra_edge_width,
extra_edge_width:-extra_edge_width]
tmp_cov = self.y_conv_2x.eval(session=self.sess,
feed_dict={
self.x: batch / 255.0,
self.y_: np.zeros((num, 2)),
self.keep_prob: 1.0,
self.l_rate: 0
})
tmp_cov = tmp_cov[0, extra_edge_width:-extra_edge_width,
extra_edge_width:-extra_edge_width,
extra_edge_width:-extra_edge_width, 1]
#tmp_res = np.max(tmp_res,axis = 1)
result[ x_start:x_start+x_len - 2*extra_edge_width,
y_start:y_start+y_len - 2*extra_edge_width,
z_start:z_start+z_len - 2*extra_edge_width]\
= tmp_cov
max_res[ x_start:x_start+x_len - 2*extra_edge_width,
y_start:y_start+y_len - 2*extra_edge_width,
z_start:z_start+z_len - 2*extra_edge_width]\
= tmp_res
te = time.time()
print(x_start, y_start, z_start, te - ts)
print(np.max(result))
print(np.max(max_res))
misc.imsave('test' + version + '_max0.png',
np.max(max_res[10:-10, 10:-10, 10:-10], axis=0) * 255)
misc.imsave('test' + version + '_max1.png',
np.max(max_res[10:-10, 10:-10, 10:-10], axis=1) * 255)
misc.imsave('test' + version + '_max2.png',
np.max(max_res[10:-10, 10:-10, 10:-10], axis=2) * 255)
misc.imsave('test' + version + '_cov0.png',
np.max(result[10:-10, 10:-10, 10:-10], axis=0) * 255)
misc.imsave('test' + version + '_cov1.png',
np.max(result[10:-10, 10:-10, 10:-10], axis=1) * 255)
misc.imsave('test' + version + '_cov2.png',
np.max(result[10:-10, 10:-10, 10:-10], axis=2) * 255)
class TransferFgTrain(object):
def __init__(self,
sess,
input_placer,
gold_placer,
learn_rate_placer,
keep_prob,
window_size=21,
model_path='./model/version_11',
image_path='./data/neuron/train',
valid_flag=False,
valid_num=2048,
nn=DNNet(),
predict_flag=True,
wd=0.001,
wdo=0.001):
self.sess = sess
self.batch_generator = ImageListBatchGenerator((window_size - 1) // 2,
image_path=image_path)
self.nn = nn
self.saver = None
self.predict_flag = predict_flag
self.fg_para_dict = dict()
self.valid_flag = valid_flag
self.window_size = window_size
if valid_flag:
self.valid = self.next_batch(valid_num)
self.get_nn(input_placer,
gold_placer,
learn_rate_placer,
keep_prob,
wd=wd,
wdo=wdo,
window_size=window_size)
print("Creation Complete\nIntialization")
init_op = tf.initialize_all_variables()
print("Start Intialization")
sess.run(init_op)
print("Intialization Done")
self.restore(model_path)
print("Saving")
#self.saver = tf.train.Saver()
#self.saver.save(self.sess,model_path + '/transfer_model.ckpt')
#print("Saved at " + model_path + '/transfer_model.ckpt')
def get_nn(self,
input_placer,
gold_placer,
learn_rate_placer,
keep_prob,
window_size,
conv_1_neuron_num=512,
conv_2_neuron_num=256,
conv_3_neuron_num=128,
conv_4_neuron_num=64,
conv_5_neuron_num=32,
wd=1e-9,
wdo=1e-9):
#create sub classes
self.x = input_placer
self.y_ = gold_placer
self.l_rate = learn_rate_placer
self.keep_prob = keep_prob
fg_dict = dict()
y_conv_combined = []
y_res_combined = []
for angle_xy in range(30, 181, 30):
for angle_xz in range(30, 181, 30):
print("Create FG NN %d %d" % (angle_xy, angle_xz))
fg_dict[(angle_xy,angle_xz)] = \
CONVDNForegroundTrain(input_placer,gold_placer,\
learn_rate_placer,keep_prob,\
angle_xy = angle_xy, \
angle_xz = angle_xz,
predict_flag = True,
valid_flag = False,
window_size = window_size)
fg_variables = tf.get_collection(
tf.GraphKeys.TRAINABLE_VARIABLES,
scope="fg_" + str(angle_xy) + '_' + str(angle_xz))
fg_dict[(angle_xy,angle_xz)].saver = \
tf.train.Saver(fg_variables)
#fg_dict[(angle_xy,angle_xz)].saver.restore(self.sess,\
# model_path + '/model_foreground_'+str(angle_xy)+'_'+str(angle_xz)+'.ckpt')
y_conv_combined.append(
tf.expand_dims(
fg_dict[(angle_xy, angle_xz)].fg_para_dict['h_conv5'],
5))
self.fg_dict = fg_dict
y_conv_combined = tf.concat(5, y_conv_combined)
dir_out = tf.reshape(
y_conv_combined,
[-1, window_size, window_size, window_size, 36 * 16])
with tf.variable_scope("transfer"):
with tf.variable_scope("transfer_layer_1"):
W_conv1 = self.nn.weight_variable(
[3, 3, 3, 36 * 16, conv_1_neuron_num], wd, 27 * 36 * 16,
27 * conv_1_neuron_num)
b_conv1 = self.nn.bias_variable([conv_1_neuron_num])
h_conv1 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(dir_out, W_conv1), b_conv1))
with tf.variable_scope("transfer_layer_2"):
W_conv2 = self.nn.weight_variable(
[3, 3, 3, conv_1_neuron_num, conv_2_neuron_num], wd,
27 * conv_1_neuron_num, 27 * conv_2_neuron_num)
b_conv2 = self.nn.bias_variable([conv_2_neuron_num])
h_conv2 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv1, W_conv2), b_conv2))
with tf.variable_scope("transfer_layer_3"):
W_conv3 = self.nn.weight_variable(
[3, 3, 3, conv_2_neuron_num, conv_3_neuron_num], wd,
27 * conv_2_neuron_num, 27 * conv_3_neuron_num)
b_conv3 = self.nn.bias_variable([conv_3_neuron_num])
h_conv3 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv2, W_conv3), b_conv3))
with tf.variable_scope("transfer_layer_4"):
W_conv4 = self.nn.weight_variable(
[3, 3, 3, conv_3_neuron_num, conv_4_neuron_num], wd,
27 * conv_3_neuron_num, 27 * conv_4_neuron_num)
b_conv4 = self.nn.bias_variable([conv_4_neuron_num])
h_conv4 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv3, W_conv4), b_conv4))
with tf.variable_scope("transfer_layer_5"):
W_conv5 = self.nn.weight_variable(
[3, 3, 3, conv_4_neuron_num, conv_5_neuron_num], wd,
27 * conv_4_neuron_num, 27 * conv_5_neuron_num)
b_conv5 = self.nn.bias_variable([conv_5_neuron_num])
h_conv5 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv4, W_conv5), b_conv5))
with tf.variable_scope("transfer_final"):
W_fc1 = self.nn.weight_variable(
[1, 1, 1, conv_5_neuron_num, 2])
b_fc1 = self.nn.bias_variable([2])
h_fc1 = tf.nn.relu(
tf.nn.bias_add(self.nn.conv2d(h_conv5, W_fc1), b_fc1))
if self.predict_flag:
self.y_conv = tf.nn.softmax(h_fc1[:, (window_size - 1) / 2,
(window_size - 1) / 2,
(window_size - 1) / 2, :])
self.y_res = tf.argmax(self.y_conv, 1)
else:
curr_shape = tf.shape(h_fc1)
reshaped_conv = tf.reshape(h_fc1, [-1, 2])
softmaxed_reshaped_conv = tf.nn.softmax(reshaped_conv)
self.y_conv = tf.reshape(softmaxed_reshaped_conv, curr_shape)
self.y_res = tf.argmax(self.y_conv, 4)
if self.valid_flag:
cross_entropy_mean = -tf.reduce_mean(self.y_ * tf.log(self.y_conv))
tf.add_to_collection('losses', cross_entropy_mean)
cross_entropy = tf.add_n(tf.get_collection('losses'),
name='total_loss')
transfer_vars = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES,
"transfer")
self.train_step = tf.train.AdamOptimizer(self.l_rate).minimize(
cross_entropy) #, var_list = transfer_vars)
correct_prediction = tf.equal(tf.argmax(self.y_conv, 1),
tf.argmax(self.y_, 1))
self.accuracy = tf.reduce_mean(tf.cast(correct_prediction,
"float"))
else:
print("Transfer Not For Training Singly")
def restore(self, model_path):
print("Restoring...")
for angle_xy in range(30, 181, 30):
for angle_xz in range(30, 181, 30):
print("Restore %d %d" % (angle_xy, angle_xz))
self.fg_dict[(angle_xy,angle_xz)].saver.restore(self.sess,\
model_path + '/model_foreground_'+str(angle_xy)+'_'+str(angle_xz)+'.ckpt')
print("Done")
def next_batch(self, batch_size):
return self.batch_generator.next_batch(batch_size)
def get_valid_set(self, valid_num):
return self.batch_generator.next_batch(valid_num)
def train(self,
epic_num=8,
loop_size=2100,
batch_size=32,
learning_rate=0.01,
thres=0.01,
keep_prob=1.0,
tao=10):
accuracies = [0.0, 0.05, 0.1, 0.15, 0.2, .25, .3, .35, .4, .45, .5]
tao_counter = 0
for epic in range(epic_num):
for i in range(loop_size):
#ts = time.time()
batch = self.next_batch(batch_size)
#te = time.time()
#print('Gen Batch: ', te-ts)
if i % 10 == 0:
print(learning_rate)
train_accuracy = 0
for v_idx in range(4):
valid = self.get_valid_set(256)
train_accuracy += self.accuracy.eval(
session=self.sess,
feed_dict={
self.x: valid[0],
self.y_: valid[1],
self.keep_prob: 1.0,
self.l_rate: learning_rate
})
train_accuracy /= 4
tao_counter += 1
#print corrects
print("%d step %d, training accuracy %g" %
(epic, i, train_accuracy))
if np.std(accuracies[-10:]) < thres or tao_counter > tao:
learning_rate = max(1e-6, learning_rate / 2)
thres /= 2
accuracies = [
0.0, 0.05, 0.1, 0.15, 0.2, .25, .3, .35, .4, .45,
.5
]
tao_counter = 0
accuracies.append(train_accuracy)
#corrects = correct_prediction.eval(feed_dict={x1: batch[0],x2: batch[1], y_: batch[2], keep_prob: 1.0,l_rate: learn_rate * (4.0-epic)})
#reinforce = append_reinforce(corrects,batch,reinforce)
#ts = time.time()
self.train_step.run(session=self.sess,
feed_dict={
self.x: batch[0],
self.y_: batch[1],
self.keep_prob: keep_prob,
self.l_rate: learning_rate
})
#te = time.time()
#print('Train: ',te-ts)
def predict(self, image, batch_size=512):
sx, sy, sz = image.shape
batches, pos, neg = sampling(image)
result = np.zeros((sx * sy * sz, 36))
max_res = np.zeros((sx * sy * sz))
counter = 0
for batch in batch_generator(batches, batch_size):
num = batch.shape[0]
#print(np.max(batch))
tmp_cov = self.y_conv_combined.eval(session=self.sess,
feed_dict={
self.x: batch / 255.0,
self.y_: np.zeros(
(num, 2)),
self.keep_prob: 1.0,
self.l_rate: 0
})
tmp_cov = np.reshape(tmp_cov[:, 1, :], (num, -1))
"""
tmp_cov = self.y_res_combined.eval(session = self.sess,
feed_dict = {
self.x:batch/255.0,
self.y_:np.zeros((num,2)),
self.keep_prob:1.0,
self.l_rate: 0
})
"""
#tmp_res = np.max(tmp_res,axis = 1)
result[counter:counter + num, :] = tmp_cov
max_res[counter:counter + num] = np.max(tmp_cov, axis=1)
counter += num
print(counter)
print(np.max(result))
result = np.reshape(result, (sx, sy, sz, 36))
max_res = np.reshape(max_res, (sx, sy, sz))
misc.imsave('test_max.png',
np.max(max_res[5:-5, 5:-5, 5:-5], axis=0) * 255)
misc.imsave('test_max.png',
np.max(max_res[5:-5, 5:-5, 5:-5], axis=1) * 255)
misc.imsave('test_max.png',
np.max(max_res[5:-5, 5:-5, 5:-5], axis=2) * 255)
for i in range(36):
misc.imsave('testr_' + str(i) + '_0.png',
np.max(result[5:-5, 5:-5, 5:-5, i], axis=0) * 255)
misc.imsave('testr_' + str(i) + '_1.png',
np.max(result[5:-5, 5:-5, 5:-5, i], axis=1) * 255)
misc.imsave('testr_' + str(i) + '_2.png',
np.max(result[5:-5, 5:-5, 5:-5, i], axis=2) * 255)
def predict_large(self, image, batch_diameter=512):
sx, sy, sz = image.shape
print(sx, sy, sz)
#batches,pos,neg = sampling(image)
result = np.zeros((sx, sy, sz, 36))
max_res = np.zeros((sx, sy, sz))
for x_start, y_start, z_start, batch in image3D_crop_generator(
image, batch_diameter):
ts = time.time()
num, x_len, y_len, z_len, _ = batch.shape
#print(batch.shape)
tmp_cov = self.y_conv_combined.eval(session=self.sess,
feed_dict={
self.x: batch / 255.0,
self.y_: np.zeros(
(num, 2)),
self.keep_prob: 1.0,
self.l_rate: 0
})
tmp_cov = tmp_cov[0, 5:-5, 5:-5, 5:-5, 1, :]
"""
tmp_cov = self.y_res_combined.eval(session = self.sess,
feed_dict = {
self.x:batch/255.0,
self.y_:np.zeros((num,2)),
self.keep_prob:1.0,
self.l_rate: 0
})
tmp_cov = tmp_cov[0,5:-5,5:-5,5:-5,:]
"""
#tmp_res = np.max(tmp_res,axis = 1)
result[x_start:x_start + x_len - 10, y_start:y_start + y_len - 10,
z_start:z_start + z_len - 10, :] = tmp_cov
te = time.time()
print(x_start, y_start, z_start, te - ts)
print(np.max(result))
max_res = np.max(result, axis=3)
misc.imsave('test_max0.png',
np.max(max_res[5:-5, 5:-5, 5:-5], axis=0) * 255)
misc.imsave('test_max1.png',
np.max(max_res[5:-5, 5:-5, 5:-5], axis=1) * 255)
misc.imsave('test_max2.png',
np.max(max_res[5:-5, 5:-5, 5:-5], axis=2) * 255)
for i in range(36):
misc.imsave('testr_' + str(i) + '_0.png',
np.max(result[5:-5, 5:-5, 5:-5, i], axis=0) * 255)
misc.imsave('testr_' + str(i) + '_1.png',
np.max(result[5:-5, 5:-5, 5:-5, i], axis=1) * 255)
misc.imsave('testr_' + str(i) + '_2.png',
np.max(result[5:-5, 5:-5, 5:-5, i], axis=2) * 255)