代码拉取完成，页面将自动刷新
Fork 仓库
加载中
确定同步？
同步操作将从 chenmingling/MRI_Deep_learning 强制同步，此操作会覆盖自 Fork 仓库以来所做的任何修改，且无法恢复！！！
确定后同步将在后台操作，完成时将刷新页面，请耐心等待。
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132
import tensorflow.contrib.slim as slim
from SRCNN_layers import *
from SRCNN_configs import *
import tensorlayer as tl
from tensorlayer.layers import *
# from loss_func import *
from EDSR_utils import *
import utils
def create_model(name, patches, n_out, is_train=False,reuse = False):
if name == 'srcnn':
return srcnn_935(patches,reuse= reuse)
elif name == 'vgg7':
return vgg7(patches)
elif name =='vgg_deconv_7':
return vgg_deconv_7(patches)
elif name == 'u_net':
return u_net(patches,n_out= n_out,reuse = reuse)
elif name == 'u_net_new':
return u_net_new(patches,n_out= n_out,reuse= reuse)
elif name == 'u_net_bn':
return u_net_bn(patches,n_out= n_out,is_train= is_train,reuse= reuse)
elif name == 'u_net_bn_new':
return u_net_bn_new(patches, n_out= n_out, is_train= is_train,reuse= reuse)
elif name == 'u_net_bn_new_2':
return u_net_bn_new_2(patches, n_out= n_out, is_train= is_train,reuse= reuse)
elif name =='srcnn_9751':
return srcnn_9751(patches,is_train)
elif name == 'SRresnet':
return SRresnet(patches, n_out= n_out, is_train= is_train,reuse= reuse)
elif name == 'EDSR':
return EDSR(patches)
elif name == 'SRDENSE':
return SRDENSE(patches, n_out= n_out, is_train= is_train,reuse= reuse)
def u_net(x, n_out=12,reuse= False): # Do I need to change n_out here ???
_, nx, ny, nz = x.get_shape().as_list()
# mean_x = tf.reduce_mean(x)
# x = x - mean_x
w_init = tf.truncated_normal_initializer(stddev=0.01)
b_init = tf.constant_initializer(value=0.0)
with tf.variable_scope("u_net",reuse = reuse):
tl.layers.set_name_reuse(reuse)
inputs = tl.layers.InputLayer(x, name='inputs')
# inputs = tf.layers.batch_normalization(inputs =inputs.outputs, axis= -1, training = True)
# inputs = tl.layers.InputLayer(inputs, name ='input_batch_layer')
conv1 = tl.layers.Conv2d(inputs, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv1_2')
pool1 = tl.layers.MaxPool2d(conv1, (2, 2), padding='SAME', name='pool1')
net1 = conv1.outputs
#variable_summaries(net1, 'net_1')
# tf.summary.image('net_1',net1[:,:,:,1],10)
conv2 = tl.layers.Conv2d(pool1, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv2_2')
pool2 = tl.layers.MaxPool2d(conv2, (2, 2), padding='SAME', name='pool2')
net2 = conv2.outputs
#variable_summaries(net2, 'net_2')
# tf.summary.image('net_2',net2,10)
conv3 = tl.layers.Conv2d(pool2, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv3_2')
pool3 = tl.layers.MaxPool2d(conv3, (2, 2), padding='SAME', name='pool3')
net3 = conv3.outputs
# variable_summaries(net3, 'net_3')
#tf.summary.image('net_3',net3,10)
conv4 = tl.layers.Conv2d(pool3, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv4_2')
pool4 = tl.layers.MaxPool2d(conv4, (2, 2), padding='SAME', name='pool4')
net4 = conv4.outputs
# variable_summaries(net4, 'net_4')
#tf.summary.image('net_4',net4,10)
conv5 = tl.layers.Conv2d(pool4, 1024, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv5_1')
conv5 = tl.layers.Conv2d(conv5, 1024, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv5_2')
net5 = conv5.outputs
# variable_summaries(net5, 'net_5')
# tf.summary.image('net_5',net5,10)
print(" * After conv: %s" % conv5.outputs)
up4 = tl.layers.DeConv2d(conv5, 512, (3, 3), out_size=(nx / 8, ny / 8), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv4')
up4 = tl.layers.ConcatLayer([up4, conv4], concat_dim=3, name='concat4')
conv4 = tl.layers.Conv2d(up4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv4_2')
netup4 = conv4.outputs
# variable_summaries(netup4, 'netup_4')
# tf.summary.image('netup_4',netup4,10)
up3 = tl.layers.DeConv2d(conv4, 256, (3, 3), out_size=(nx / 4, ny / 4), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv3')
up3 = tl.layers.ConcatLayer([up3, conv3], concat_dim=3, name='concat3')
conv3 = tl.layers.Conv2d(up3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv3_2')
netup3 = conv3.outputs
# variable_summaries(netup3, 'netup_3')
# tf.summary.image('netup_3',netup3,10)
up2 = tl.layers.DeConv2d(conv3, 128, (3, 3), out_size=(nx / 2, ny / 2), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv2')
up2 = tl.layers.ConcatLayer([up2, conv2], concat_dim=3, name='concat2')
conv2 = tl.layers.Conv2d(up2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv2_2')
netup2 = conv2.outputs
# variable_summaries(netup2, 'netup_2')
# tf.summary.image('netup_2',netup2,10)
up1 = tl.layers.DeConv2d(conv2, 64, (3, 3), out_size=(nx / 1, ny / 1), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv1')
up1 = tl.layers.ConcatLayer([up1, conv1], concat_dim=3, name='concat1')
conv1 = tl.layers.Conv2d(up1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv1_2')
netup1 = conv1.outputs
# variable_summaries(netup1, 'netup_1')
# tf.summary.image('netup_1',netup1,10)
conv1 = tl.layers.Conv2d(conv1, n_out, (1, 1), act=None, name='uconv1')
# tf.summary.image('final', tf.cast(conv1.outputs,tf.uint8), 1)
# tf.summary.image('final', conv1.outputs, 10)
conv1 = tf.add(conv1.outputs, inputs.outputs)
# print(" * Output: %s" % conv1.outputs)
#outputs = tl.act.pixel_wise_softmax(conv1.outputs)
return conv1
def u_net_bn(x, n_out=12,is_train = False, reuse=False): # Do I need to change n_out here ???
_, nx, ny, nz = x.get_shape().as_list()
# mean_x = tf.reduce_mean(x)
# x = x - mean_x
w_init = tf.truncated_normal_initializer(stddev=0.01)
b_init = tf.constant_initializer(value=0.0)
gamma_init = tf.random_normal_initializer(1, 0.02)
with tf.variable_scope("u_net", reuse=reuse):
tl.layers.set_name_reuse(reuse)
inputs = tl.layers.InputLayer(x, name='inputs')
# inputs = tf.layers.batch_normalization(inputs =inputs.outputs, axis= -1, training = True)
# inputs = tl.layers.InputLayer(inputs, name ='input_batch_layer')
conv1 = tl.layers.Conv2d(inputs, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv1_2')
conv1 = BatchNormLayer(conv1, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn1')
pool1 = tl.layers.MaxPool2d(conv1, (2, 2), padding='SAME', name='pool1')
net1 = conv1.outputs
# variable_summaries(net1, 'net_1')
# tf.summary.image('net_1',net1[:,:,:,1],10)
conv2 = tl.layers.Conv2d(pool1, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv2_2')
conv2 = BatchNormLayer(conv2, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn2')
pool2 = tl.layers.MaxPool2d(conv2, (2, 2), padding='SAME', name='pool2')
net2 = conv2.outputs
# variable_summaries(net2, 'net_2')
# tf.summary.image('net_2',net2,10)
conv3 = tl.layers.Conv2d(pool2, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv3_2')
conv3 = BatchNormLayer(conv3, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn3')
pool3 = tl.layers.MaxPool2d(conv3, (2, 2), padding='SAME', name='pool3')
net3 = conv3.outputs
# variable_summaries(net3, 'net_3')
# tf.summary.image('net_3',net3,10)
conv4 = tl.layers.Conv2d(pool3, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv4_2')
conv4 = BatchNormLayer(conv4, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn4')
pool4 = tl.layers.MaxPool2d(conv4, (2, 2), padding='SAME', name='pool4')
net4 = conv4.outputs
# variable_summaries(net4, 'net_4')
# tf.summary.image('net_4',net4,10)
conv5 = tl.layers.Conv2d(pool4, 1024, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv5_1')
conv5 = tl.layers.Conv2d(conv5, 1024, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv5_2')
conv5 = BatchNormLayer(conv5, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn5')
net5 = conv5.outputs
# variable_summaries(net5, 'net_5')
# tf.summary.image('net_5',net5,10)
print(" * After conv: %s" % conv5.outputs)
up4 = tl.layers.DeConv2d(conv5, 512, (3, 3), out_size=(nx / 8, ny / 8), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv4')
up4 = tl.layers.ConcatLayer([up4, conv4], concat_dim=3, name='concat4')
conv4 = tl.layers.Conv2d(up4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv4_2')
conv4 = BatchNormLayer(conv4, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='ucov_bn4')
netup4 = conv4.outputs
# variable_summaries(netup4, 'netup_4')
# tf.summary.image('netup_4',netup4,10)
up3 = tl.layers.DeConv2d(conv4, 256, (3, 3), out_size=(nx / 4, ny / 4), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv3')
up3 = tl.layers.ConcatLayer([up3, conv3], concat_dim=3, name='concat3')
conv3 = tl.layers.Conv2d(up3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv3_2')
conv3 = BatchNormLayer(conv3, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='ucov_bn3')
netup3 = conv3.outputs
# variable_summaries(netup3, 'netup_3')
# tf.summary.image('netup_3',netup3,10)
up2 = tl.layers.DeConv2d(conv3, 128, (3, 3), out_size=(nx / 2, ny / 2), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv2')
up2 = tl.layers.ConcatLayer([up2, conv2], concat_dim=3, name='concat2')
conv2 = tl.layers.Conv2d(up2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv2_2')
conv2 = BatchNormLayer(conv2, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='ucov_bn2')
netup2 = conv2.outputs
# variable_summaries(netup2, 'netup_2')
# tf.summary.image('netup_2',netup2,10)
up1 = tl.layers.DeConv2d(conv2, 64, (3, 3), out_size=(nx / 1, ny / 1), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv1')
up1 = tl.layers.ConcatLayer([up1, conv1], concat_dim=3, name='concat1')
conv1 = tl.layers.Conv2d(up1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv1_2')
conv1 = BatchNormLayer(conv1, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='ucov_bn1')
netup1 = conv1.outputs
# variable_summaries(netup1, 'netup_1')
# tf.summary.image('netup_1',netup1,10)
conv1 = tl.layers.Conv2d(conv1, n_out, (1, 1), act=None, name='uconv1')
# tf.summary.image('final', tf.cast(conv1.outputs,tf.uint8), 1)
# tf.summary.image('final', conv1.outputs, 10)
conv1 = tf.add(conv1.outputs, inputs.outputs)
# print(" * Output: %s" % conv1.outputs)
# outputs = tl.act.pixel_wise_softmax(conv1.outputs)
return conv1
def u_net_bn_new(x, n_out=12, is_train=False, reuse=False): # Do I need to change n_out here ???
_, nx, ny, nz = x.get_shape().as_list()
# mean_x = tf.reduce_mean(x)
# x = x - mean_x
w_init = tf.truncated_normal_initializer(stddev=0.01)
b_init = tf.constant_initializer(value=0.0)
gamma_init = tf.random_normal_initializer(1, 0.02)
with tf.variable_scope("u_net", reuse=reuse):
tl.layers.set_name_reuse(reuse)
inputs = tl.layers.InputLayer(x, name='inputs')
# inputs = tf.layers.batch_normalization(inputs =inputs.outputs, axis= -1, training = True)
# inputs = tl.layers.InputLayer(inputs, name ='input_batch_layer')
conv1 = tl.layers.Conv2d(inputs, 64, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv1_1')
conv1 = BatchNormLayer(conv1, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv1_2')
conv1 = BatchNormLayer(conv1, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn1_2')
pool1 = tl.layers.MaxPool2d(conv1, (2, 2), padding='SAME', name='pool1')
net1 = conv1.outputs
# variable_summaries(net1, 'net_1')
# tf.summary.image('net_1',net1[:,:,:,1],10)
conv2 = tl.layers.Conv2d(pool1, 128, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv2_1')
conv2 = BatchNormLayer(conv2, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv2_2')
conv2 = BatchNormLayer(conv2, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn2_2')
pool2 = tl.layers.MaxPool2d(conv2, (2, 2), padding='SAME', name='pool2')
net2 = conv2.outputs
# variable_summaries(net2, 'net_2')
# tf.summary.image('net_2',net2,10)
conv3 = tl.layers.Conv2d(pool2, 256, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv3_1')
conv3 = BatchNormLayer(conv3, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv3_2')
conv3 = BatchNormLayer(conv3, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn3_2')
pool3 = tl.layers.MaxPool2d(conv3, (2, 2), padding='SAME', name='pool3')
net3 = conv3.outputs
# variable_summaries(net3, 'net_3')
# tf.summary.image('net_3',net3,10)
conv4 = tl.layers.Conv2d(pool3, 512, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv4_1')
conv4 = BatchNormLayer(conv4, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv4_2')
conv4 = BatchNormLayer(conv4, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn4_2')
pool4 = tl.layers.MaxPool2d(conv4, (2, 2), padding='SAME', name='pool4')
net4 = conv4.outputs
# variable_summaries(net4, 'net_4')
# tf.summary.image('net_4',net4,10)
conv5 = tl.layers.Conv2d(pool4, 1024, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv5_1')
conv5 = BatchNormLayer(conv5, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn5_1')
conv5 = tl.layers.Conv2d(conv5, 1024, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='conv5_2')
conv5 = BatchNormLayer(conv5, act=lambda x: tl.act.lrelu(x, 0.2), is_train=is_train, gamma_init=gamma_init,
name='bn5_2')
net5 = conv5.outputs
# variable_summaries(net5, 'net_5')
# tf.summary.image('net_5',net5,10)
print(" * After conv: %s" % conv5.outputs)
up4 = tl.layers.DeConv2d(conv5, 512, (3, 3), out_size=(nx / 8, ny / 8), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv4')
up4 = BatchNormLayer(up4, act=tf.nn.relu, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn4_1')
up4 = tl.layers.ConcatLayer([up4, conv4], concat_dim=3, name='concat4')
conv4 = tl.layers.Conv2d(up4, 512, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv4_2')
conv4 = BatchNormLayer(conv4, act=tf.nn.relu, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn4_2')
netup4 = conv4.outputs
# variable_summaries(netup4, 'netup_4')
# tf.summary.image('netup_4',netup4,10)
up3 = tl.layers.DeConv2d(conv4, 256, (3, 3), out_size=(nx / 4, ny / 4), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv3')
up3 = BatchNormLayer(up3, act=tf.nn.relu, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn3_1')
up3 = tl.layers.ConcatLayer([up3, conv3], concat_dim=3, name='concat3')
conv3 = tl.layers.Conv2d(up3, 256, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv3_2')
conv3 = BatchNormLayer(conv3, act=tf.nn.relu, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn3_2')
netup3 = conv3.outputs
# variable_summaries(netup3, 'netup_3')
# tf.summary.image('netup_3',netup3,10)
up2 = tl.layers.DeConv2d(conv3, 128, (3, 3), out_size=(nx / 2, ny / 2), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv2')
up2 = BatchNormLayer(up2, act=tf.nn.relu, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn2_1')
up2 = tl.layers.ConcatLayer([up2, conv2], concat_dim=3, name='concat2')
conv2 = tl.layers.Conv2d(up2, 128, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv2_2')
conv2 = BatchNormLayer(conv2, act=tf.nn.relu, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn2_2')
netup2 = conv2.outputs
# variable_summaries(netup2, 'netup_2')
# tf.summary.image('netup_2',netup2,10)
up1 = tl.layers.DeConv2d(conv2, 64, (3, 3), out_size=(nx / 1, ny / 1), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv1')
up1 = BatchNormLayer(up1, act=tf.nn.relu, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn1_1')
up1 = tl.layers.ConcatLayer([up1, conv1], concat_dim=3, name='concat1')
conv1 = tl.layers.Conv2d(up1, 64, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv1_2')
conv1 = BatchNormLayer(conv1, act=tf.nn.relu, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn1_2')
netup1 = conv1.outputs
# variable_summaries(netup1, 'netup_1')
# tf.summary.image('netup_1',netup1,10)
# conv1 = tl.layers.Conv2d(conv1, n_out, (1, 1), act=tf.nn.tanh, name='uconv1')
# tf.summary.image('final', tf.cast(conv1.outputs,tf.uint8), 1)
conv1 = tl.layers.Conv2d(conv1, 32, (3, 3), act=tf.nn.relu, name='uconv1')
conv1 = tl.layers.Conv2d(conv1, n_out, (1, 1), act=None, name='uconv1_new')
# tf.summary.image('final', conv1.outputs, 10)
conv1 = tf.add(conv1.outputs, inputs.outputs)
# print(" * Output: %s" % conv1.outputs)
# outputs = tl.act.pixel_wise_softmax(conv1.outputs)
return conv1
def u_net_bn_new_2(x, n_out=12, is_train=False, reuse=False):
batch_size, nx, ny, nz = x.get_shape().as_list()
# mean_x = tf.reduce_mean(x)
# x = x - mean_x
w_init = tf.truncated_normal_initializer(stddev=0.01)
b_init = tf.constant_initializer(value=0.0)
gamma_init = tf.random_normal_initializer(1, 0.02)
with tf.variable_scope("u_net", reuse=reuse):
tl.layers.set_name_reuse(reuse)
inputs = tl.layers.InputLayer(x, name='inputs')
# inputs = tf.layers.batch_normalization(inputs =inputs.outputs, axis= -1, training = True)
# inputs = tl.layers.InputLayer(inputs, name ='input_batch_layer')
conv1 = tl.layers.Conv2d(inputs, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv1_2')
conv1 = BatchNormLayer(conv1, is_train=is_train, gamma_init=gamma_init,
name='bn1')
pool1 = tl.layers.MaxPool2d(conv1, (2, 2), padding='SAME', name='pool1')
net1 = conv1.outputs
# variable_summaries(net1, 'net_1')
# tf.summary.image('net_1',net1[:,:,:,1],10)
conv2 = tl.layers.Conv2d(pool1, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv2_2')
conv2 = BatchNormLayer(conv2, is_train=is_train, gamma_init=gamma_init,
name='bn2')
pool2 = tl.layers.MaxPool2d(conv2, (2, 2), padding='SAME', name='pool2')
net2 = conv2.outputs
# variable_summaries(net2, 'net_2')
# tf.summary.image('net_2',net2,10)
conv3 = tl.layers.Conv2d(pool2, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv3_2')
conv3 = BatchNormLayer(conv3, is_train=is_train, gamma_init=gamma_init,
name='bn3')
pool3 = tl.layers.MaxPool2d(conv3, (2, 2), padding='SAME', name='pool3')
net3 = conv3.outputs
# variable_summaries(net3, 'net_3')
# tf.summary.image('net_3',net3,10)
conv4 = tl.layers.Conv2d(pool3, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv4_2')
conv4 = BatchNormLayer(conv4, is_train=is_train, gamma_init=gamma_init,
name='bn4')
pool4 = tl.layers.MaxPool2d(conv4, (2, 2), padding='SAME', name='pool4')
net4 = conv4.outputs
# variable_summaries(net4, 'net_4')
# tf.summary.image('net_4',net4,10)
conv5 = tl.layers.Conv2d(pool4, 1024, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv5_1')
conv5 = tl.layers.Conv2d(conv5, 1024, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='conv5_2')
conv5 = BatchNormLayer(conv5, is_train=is_train, gamma_init=gamma_init,
name='bn5')
net5 = conv5.outputs
# variable_summaries(net5, 'net_5')
# tf.summary.image('net_5',net5,10)
print(" * After conv: %s" % conv5.outputs)
up4 = tl.layers.DeConv2d(conv5, 512, (3, 3), out_size=[tf.to_int32(tf.shape(x)[1] / 8),tf.to_int32(tf.shape(x)[2]/ 8)], strides=(2, 2),
padding='SAME', act=tf.nn.relu, W_init=w_init, b_init=b_init, name='deconv4')
up4 = BatchNormLayer(up4, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn4_1')
up4 = tl.layers.ConcatLayer([up4, conv4], concat_dim=3, name='concat4')
conv4 = tl.layers.Conv2d(up4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv4_2')
conv4 = BatchNormLayer(conv4, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn4_2')
netup4 = conv4.outputs
# variable_summaries(netup4, 'netup_4')
# tf.summary.image('netup_4',netup4,10)
up3 = tl.layers.DeConv2d(conv4, 256, (3, 3), out_size=[tf.to_int32(tf.shape(x)[1] / 4),tf.to_int32(tf.shape(x)[2]/ 4)], strides=(2, 2),
padding='SAME', act=tf.nn.relu, W_init=w_init, b_init=b_init, name='deconv3')
up3 = BatchNormLayer(up3, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn3_1')
up3 = tl.layers.ConcatLayer([up3, conv3], concat_dim=3, name='concat3')
conv3 = tl.layers.Conv2d(up3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv3_2')
conv3 = BatchNormLayer(conv3, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn3_2')
netup3 = conv3.outputs
# variable_summaries(netup3, 'netup_3')
# tf.summary.image('netup_3',netup3,10)
up2 = tl.layers.DeConv2d(conv3, 128, (3, 3), out_size=[tf.to_int32(tf.shape(x)[1] / 2),tf.to_int32(tf.shape(x)[2]/ 2)], strides=(2, 2),
padding='SAME', act=tf.nn.relu, W_init=w_init, b_init=b_init, name='deconv2')
up2 = BatchNormLayer(up2, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn2_1')
up2 = tl.layers.ConcatLayer([up2, conv2], concat_dim=3, name='concat2')
conv2 = tl.layers.Conv2d(up2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv2_2')
conv2 = BatchNormLayer(conv2, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn2_2')
netup2 = conv2.outputs
# variable_summaries(netup2, 'netup_2')
# tf.summary.image('netup_2',netup2,10)
up1 = tl.layers.DeConv2d(conv2, 64, (3, 3), out_size=[tf.to_int32(tf.shape(x)[1] ),tf.to_int32(tf.shape(x)[2])], strides=(2, 2),
padding='SAME', act=tf.nn.relu, W_init=w_init, b_init=b_init, name='deconv1')
up1 = BatchNormLayer(up1, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn1_1')
up1 = tl.layers.ConcatLayer([up1, conv1], concat_dim=3, name='concat1')
conv1 = tl.layers.Conv2d(up1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='uconv1_2')
conv1 = BatchNormLayer(conv1, is_train=is_train, gamma_init=gamma_init,
name='ucov_bn1_2')
netup1 = conv1.outputs
# variable_summaries(netup1, 'netup_1')
# tf.summary.image('netup_1',netup1,10)
conv1 = tl.layers.Conv2d(conv1, n_out, (1, 1), act=None, name='uconv1')
# tf.summary.image('final', tf.cast(conv1.outputs,tf.uint8), 1)
# tf.summary.image('final', conv1.outputs, 10)
conv1 = tf.add(conv1.outputs, inputs.outputs,name = 'output')
# input = inputs.outputs
######## -------------------------Data fidelity--------------------------------##########
# for contrast in range(n_out):
# k_conv3 = utils.Fourier(conv1[:,:,:,contrast], separate_complex=False)
# mask = np.ones((batch_size, nx, ny))
# mask[:,:, 1:ny:3] = 0
# mask = np.fft.ifftshift(mask)
# # convert to complex tf tensor
# DEFAULT_MAKS_TF = tf.cast(tf.constant(mask), tf.float32)
# DEFAULT_MAKS_TF_c = tf.cast(DEFAULT_MAKS_TF, tf.complex64)
# k_patches = utils.Fourier(input[:,:,:,contrast], separate_complex=False)
# k_space = k_conv3 * DEFAULT_MAKS_TF_c + k_patches*(1-DEFAULT_MAKS_TF_c)
# out = tf.ifft2d(k_space)
# out = tf.abs(out)
# out = tf.reshape(out, [batch_size, nx, ny, 1])
# if contrast == 0 :
# final_output = out
# else:
# final_output = tf.concat([final_output,out],3)
########-------------------------end------------------------------------###########3
# print(" * Output: %s" % conv1.outputs)
# outputs = tl.act.pixel_wise_softmax(conv1.outputs)
return conv1
def u_net_new(x, n_out=12, reuse = False): # Do I need to change n_out here ???
_, nx, ny, nz = x.get_shape().as_list()
w_init = tf.truncated_normal_initializer(stddev=0.01)
b_init = tf.constant_initializer(value=0.0)
# gamma_init = tf.random_normal_initializer(1, 0.02)
with tf.variable_scope("u_net", reuse= reuse):
tl.layers.set_name_reuse(reuse)
inputs = tl.layers.InputLayer(x, name ='inputs')
conv1 = tl.layers.Conv2d(inputs, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv1_2')
pool1 = tl.layers.MaxPool2d(conv1, (2, 2), padding='SAME', name='pool1')
# 第二层
conv2 = tl.layers.Conv2d(pool1, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv2_2')
pool2 = tl.layers.MaxPool2d(conv2, (2, 2), padding='SAME', name='pool2')
#第三层
conv3 = tl.layers.Conv2d(pool2, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv3_2')
pool3 = tl.layers.MaxPool2d(conv3, (2, 2), padding='SAME', name='pool3')
#第四层
conv4 = tl.layers.Conv2d(pool3, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv4_2')
pool4 = tl.layers.MaxPool2d(conv4, (2, 2), padding='SAME', name='pool4')
# # 第五层
conv5 = tl.layers.Conv2d(pool4, 1024, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv5_1')
conv5 = tl.layers.Conv2d(conv5, 1024, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='conv5_2')
up4 = tl.layers.DeConv2d(conv5, 512, (3, 3), out_size=(nx / 8, ny / 8), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv4')
up4 = tl.layers.ConcatLayer([up4, conv4], concat_dim=3, name='concat4')
conv4 = tl.layers.Conv2d(up4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv4_1')
conv4 = tl.layers.Conv2d(conv4, 512, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv4_2')
#第七层
up3 = tl.layers.DeConv2d(conv4, 256, (3, 3), out_size=(nx / 4, ny / 4), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv3')
up3 = tl.layers.ConcatLayer([up3, conv3], concat_dim=3, name='concat3')
conv3 = tl.layers.Conv2d(up3, 256, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init, name='uconv3_1')
conv3 = tl.layers.Conv2d(conv3, 256, (3, 3), act=None, padding='SAME', W_init=w_init, b_init=b_init, name='uconv3_2')
up2 = tl.layers.DeConv2d(conv3, 128, (3, 3), out_size=(nx / 2, ny / 2), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv2')
up2 = tl.layers.ConcatLayer([up2, conv2], concat_dim=3, name='concat2')
conv2 = tl.layers.Conv2d(up2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv2_1')
conv2 = tl.layers.Conv2d(conv2, 128, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv2_2')
# ###会引起报错的代码
# netup2 = conv2.outputs
# variable_summaries(netup2, 'netup_2')
up1 = tl.layers.DeConv2d(conv2, 64, (3, 3), out_size=(nx / 1, ny / 1), strides=(2, 2),
padding='SAME', act=None, W_init=w_init, b_init=b_init, name='deconv1')
up1 = tl.layers.ConcatLayer([up1, conv1], concat_dim=3, name='concat1')
conv1 = tl.layers.Conv2d(up1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv1_1')
conv1 = tl.layers.Conv2d(conv1, 64, (3, 3), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name='uconv1_2')
# conv1 = BatchNormLayer(conv1, act = lambda x: tl.act.lrelu(x, 0.2), is_train= is_train, gamma_init=gamma_init, name= 'uconv1_bn1')
conv1 = tl.layers.Conv2d(conv1, n_out, (1, 1), act=None, name='uconv1')
# conv1 = tf.layers.batch_normalization(inputs =conv1.outputs, axis= -1, training = True)
# conv1 = tl.layers.InputLayer(conv1, name ='batch_layer9')
# netup1 = conv1.outputs
# variable_summaries(netup1, 'netup_1')
# tf.summary.image('netup_1',netup1,10)
# print(" * Output: %s" % conv1.outputs)
#outputs = tl.act.pixel_wise_softmax(conv1.outputs)
return conv1.outputs
#### DENSE block
# def DesWBH(desBlock_layer, filter_size, outlayer):
# weightsH = {}
# biasesH = {}
# fs = filter_size
# for i in range(1, outlayer+1):
# for j in range(1, desBlock_layer+1):
# if j is 1:
# weightsH.update({'w_H_%d_%d' % (i, j): tf.Variable(tf.random_normal([fs, fs, 6, 6], stddev=np.sqrt(2.0/9/16)), name='w_H_%d_%d' % (i, j))})
# else:
# weightsH.update({'w_H_%d_%d' % (i, j): tf.Variable(tf.random_normal([fs, fs, 16* (j-1), 16], stddev=np.sqrt(2.0/9/(16 * (j-1)))), name='w_H_%d_%d' % (i, j))})
# biasesH.update({'b_H_%d_%d' % (i, j): tf.Variable(tf.zeros([16], name='b_H_%d_%d' % (i, j)))})
# return weightsH, biasesH
#
#
# def Concatenation(layers):
# return tf.concat(layers, axis=3)
#
#
# def SkipConnect(conv):
# skipconv = list()
# for i in conv:
# x = Concatenation(i)
# skipconv.append(x)
# return skipconv
# def desBlock(input, desBlock_layer, outlayer, filter_size=3 ):
#
# des_block_H =8
# des_block_ALL=8
# weight_block, biases_block = DesWBH(des_block_H, 3, des_block_ALL)
# w_init = tf.truncated_normal_initializer(stddev=np.sqrt(2.0/9/16))
# b_init = tf.constant_initializer(value=0.0)
#
# nextlayer = input
# conv = list()
# for i in range(1, outlayer+1):
# conv_in = list()
# for j in range(1, desBlock_layer+1):
# # The first conv need connect with low level layer
# if j is 1:
#
# x = tf.nn.conv2d(nextlayer, weight_block['w_H_%d_%d' %(i, j)], strides=[1,1,1,1], padding='SAME') + biases_block['b_H_%d_%d' % (i, j)]
# x = tf.nn.relu(x)
# conv_in.append(x)
# else:
# x = Concatenation(conv_in)
# x = tf.nn.conv2d(x, weight_block['w_H_%d_%d' % (i, j)], strides=[1,1,1,1], padding='SAME')+ biases_block['b_H_%d_%d' % (i, j)]
# x = tf.nn.relu(x)
# conv_in.append(x)
#
# nextlayer = conv_in[-1]
# conv.append(conv_in)
# return conv
# def bot_layer(input_layer):
# # Bottleneck layer
# des_block_H =8
# des_block_ALL=8
# allfeature = 16 * des_block_H * des_block_ALL + 16
# bot_weight = tf.Variable(tf.random_normal([1, 1, allfeature, 256], stddev=np.sqrt(2.0 / 1 / allfeature)),
# name='w_bot')
# bot_biases = tf.Variable(tf.zeros([256], name='b_bot'))
# x = tf.nn.conv2d(input_layer, bot_weight, strides=[1,1,1,1], padding='SAME') + bot_biases
# x = tf.nn.relu(x)
# return x
# def reconv_layer( input_layer, n_out):
# reconv_weight = tf.Variable(tf.random_normal([3, 3, 256, n_out], stddev=np.sqrt(2.0 / 9 / 256)),
# name='w_reconv')
# reconv_biases = tf.Variable(tf.zeros([6], name='b_reconv'))
# x = tf.nn.conv2d(input_layer, reconv_weight, strides=[1,1,1,1], padding='SAME') + reconv_biases
# return x
# def SRDENSE(patches, n_out, is_train, reuse):
# with tf.variable_scope("SRDENSE", reuse=reuse):
# tl.layers.set_name_reuse(reuse)
# x = desBlock(patches, 8, 8, filter_size=3)
# # NOTE: Cocate all dense block
#
# x = SkipConnect(x)
# x.append(patches)
# x = Concatenation(x)
# x = bot_layer(x)
# x = reconv_layer(x, n_out)
#
# return x
#### DENSE block
def DesWBH(desBlock_layer, filter_size, outlayer):
weightsH = {}
biasesH = {}
fs = filter_size
for i in range(1, outlayer+1):
for j in range(1, desBlock_layer+1):
if j is 1:
weightsH.update({'w_H_%d_%d' % (i, j): tf.Variable(tf.random_normal([fs, fs, 16, 16], stddev=np.sqrt(2.0/9/16)), name='w_H_%d_%d' % (i, j))})
else:
weightsH.update({'w_H_%d_%d' % (i, j): tf.Variable(tf.random_normal([fs, fs, 16* (j-1), 16], stddev=np.sqrt(2.0/9/(16 * (j-1)))), name='w_H_%d_%d' % (i, j))})
biasesH.update({'b_H_%d_%d' % (i, j): tf.Variable(tf.zeros([16], name='b_H_%d_%d' % (i, j)))})
return weightsH, biasesH
# def Concatenation(layers,name):
# return tl.layers.ConcatLayer(layers, concat_dim=3,name=name)
def Concatenation(layers,name):
return tl.layers.ConcatLayer(layers, concat_dim=3,name=name)
def SkipConnect(conv):
skipconv = list()
# x = Concatenation(conv,name='skip')
# skipconv.append(x)
k=0
for i in conv:
k +=1
x = Concatenation(i,name = 'skip'+str(k))
skipconv.append(x)
return skipconv
def desBlock(input, desBlock_layer, outlayer, filter_size=3, is_train= False ):
# w_init = tf.truncated_normal_initializer(stddev=np.sqrt(2/9/16))
w_init = tf.random_normal_initializer(stddev=0.01)
b_init = tf.constant_initializer(value=0.0)
gamma_init = tf.random_normal_initializer(1, 0.02)
nextlayer = input
conv = list()
for i in range(1, outlayer+1):
conv_in = list()
for j in range(1, desBlock_layer+1):
# The first conv need connect with low level layer
if j is 1:
x = tl.layers.Conv2d(nextlayer,16,(3,3),act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name ='desblock_'+str(i)+'_'+str(j) )
# x = BatchNormLayer(x, is_train=is_train, gamma_init=gamma_init,
# name='desblock_'+str(i)+'_'+str(j)+'bn')
conv_in.append(x)
else:
x = Concatenation(conv_in,name = 'concat'+str(i)+'_'+str(j) )
x = tl.layers.Conv2d(x,16,(3,3),act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init, name ='desblock_'+str(i)+'_'+str(j))
# x = BatchNormLayer(x, is_train=is_train, gamma_init=gamma_init,
# name='desblock_'+str(i)+'_'+str(j)+'bn')
conv_in.append(x)
nextlayer = conv_in[-1]
conv.append(conv_in)
return conv
def bot_layer(input_layer,is_train):
# Bottleneck layer
des_block_H =8
des_block_ALL=8
allfeature = 16 * des_block_H * des_block_ALL + 16
# bot_weight = tf.random_normal_initializer(stddev=np.sqrt(2/9/allfeature))
bot_weight = tf.random_normal_initializer(stddev=0.01)
bot_biases = tf.constant_initializer(value=0.0)
gamma_init = tf.random_normal_initializer(1, 0.02)
x = tl.layers.Conv2d(input_layer, 256, (3, 3), act=tf.nn.relu, padding='SAME', W_init=bot_weight, b_init=bot_biases,name = 'bot_layer')
# x = BatchNormLayer(x, is_train=is_train, gamma_init=gamma_init,
# name='bot_layer_' + 'bn')
return x
def reconv_layer( input_layer, n_out,is_train):
# reconv_weight = tf.random_normal_initializer(stddev=np.sqrt(2/9/256))
reconv_weight = tf.random_normal_initializer(stddev=0.01)
reconv_biases = tf.constant_initializer(value=0.0)
gamma_init = tf.random_normal_initializer(1, 0.02)
x = tl.layers.Conv2d(input_layer, n_out, (3, 3), act=None, padding='SAME', W_init=reconv_weight, b_init=reconv_biases, name='recon_layer')
# x = BatchNormLayer(x, is_train=is_train, gamma_init=gamma_init,
# name='reconv_layer_' + 'bn')
return x
def SRDENSE(patches, n_out, is_train, reuse):
with tf.variable_scope("SRDENSE", reuse=reuse):
tl.layers.set_name_reuse(reuse)
inputs = tl.layers.InputLayer(patches, name='inputs')
x = desBlock(inputs, 8, 8, filter_size=3,is_train = is_train)
# x = Concatenation(x, name='concat')
x = SkipConnect(x)
x.append(inputs)
# NOTE: Cocate all dense block
x = Concatenation(x,name = 'con_final')
x = bot_layer(x,is_train)
x = reconv_layer(x, n_out,is_train)
final = tf.add(x.outputs, inputs.outputs)
return final
def srcnn_9751(patches,name ='zqq'):
with tf.variable_scope(name):
#upscaled_patches = tf.image.resize_bicubic(patches, [INPUT_SIZE, INPUT_SIZE], True)
conv1 = conv2d(patches, 9, 9, 64, padding='SAME', name='conv1')
relu1 = relu(conv1, name='relu1')
dd = tf.transpose(relu1, perm=[3,1,2,0])
# tf.summary.image('conv1', dd, 10)
conv2 = conv2d(relu1, 7, 7, 32, padding='SAME', name='conv2')
relu2 = relu(conv2, name='relu2')
dd = tf.transpose(relu2, perm=[3,1,2,0])
# tf.summary.image('conv2', dd, 10)
conv3 = conv2d(relu2, 1, 1, 16, padding='SAME', name='conv2')
relu3 = relu(conv3, name='relu3')
return conv2d(relu3, 5, 5, NUM_CHENNELS, padding='SAME', name='conv3')
def srcnn_935(patches, name='srcnn',reuse = False):
batch_size, nx, ny, nz = patches.get_shape().as_list()
w_init = tf.truncated_normal_initializer(stddev=0.01)
b_init = tf.constant_initializer(value=0.0)
with tf.variable_scope(name,reuse= reuse):
tl.layers.set_name_reuse(reuse)
inputs = tl.layers.InputLayer(patches,name = 'inputs')
conv1 = tl.layers.Conv2d(inputs,64,(9,9),act = tf.nn.relu, padding='SAME',W_init=w_init,b_init= b_init, name = 'conv1')
conv2 = tl.layers.Conv2d(conv1,32,(3,3),act = tf.nn.relu, padding='SAME',W_init=w_init,b_init= b_init, name = 'conv2')
conv3 = tl.layers.Conv2d(conv2,1,(5,5),act = None, padding='SAME',W_init=w_init,b_init= b_init, name = 'conv3')
#upscaled_patches = tf.image.resize_bicubic(patches, [INPUT_SIZE, INPUT_SIZE], True)
######## Data fidelity
k_conv3 = utils.Fourier(conv3.outputs,separate_complex = False)
mask = np.ones((batch_size, nx, ny))
mask[:,np.array(nx/4+1,dtype= 'int8'):np.array(nx*3/4,dtype= 'int8'),:] = 0
mask = np.fft.ifftshift(mask)
# convert to complex tf tensor
DEFAULT_MAKS_TF = tf.cast(tf.constant(mask), tf.float32)
DEFAULT_MAKS_TF_c = tf.cast(DEFAULT_MAKS_TF, tf.complex64)
k_patches = utils.Fourier(patches,separate_complex=False)
k_space = k_conv3 * DEFAULT_MAKS_TF_c + k_patches
out = tf.ifft2d(k_space)
out = tf.abs(out)
out = tf.reshape(out,[batch_size,nx,ny,1])
# k_conv3[:,nx/4+1:nx*3/4,ny/4+1:ny*3/4,:] = k_patches[:,nx/4+1:nx*3/4,ny/4+1:ny*3/4,:]
return out
def SRresnet(t_image, n_out= 12,is_train=False, reuse=False):
w_init = tf.random_normal_initializer(stddev=0.02)
b_init = None # tf.constant_initializer(value=0.0)
g_init = tf.random_normal_initializer(1., 0.02)
with tf.variable_scope("SRGAN_g", reuse=reuse) as vs:
tl.layers.set_name_reuse(reuse)
Input = InputLayer(t_image, name='in')
n = Conv2d(Input, 64, (3, 3), (1, 1), act=tf.nn.relu, padding='SAME', W_init=w_init, name='n64s1/c')
temp = n
# B residual blocks
for i in range(32):
nn = Conv2d(n, 64, (3, 3), (1, 1), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='n64s1/c1/%s' % i)
nn = BatchNormLayer(nn, act=tf.nn.relu, is_train=is_train, gamma_init=g_init, name='n64s1/b1/%s' % i)
nn = Conv2d(nn, 64, (3, 3), (1, 1), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='n64s1/c2/%s' % i)
nn = BatchNormLayer(nn, is_train=is_train, gamma_init=g_init, name='n64s1/b2/%s' % i)
nn = ElementwiseLayer([n, nn], tf.add, 'b_residual_add/%s' % i)
n = nn
n = Conv2d(n, 64, (3, 3), (1, 1), act=None, padding='SAME', W_init=w_init, b_init=b_init, name='n64s1/c/m')
n = BatchNormLayer(n, is_train=is_train, gamma_init=g_init, name='n64s1/b/m')
n = ElementwiseLayer([n, temp], tf.add, 'add3')
# B residual blacks end
n = Conv2d(n, 256, (3, 3), (1, 1), act=tf.nn.relu, padding='SAME', W_init=w_init, name='n256s1/1')
# n = SubpixelConv2d(n, scale=1, n_out_channel=None, act=tf.nn.relu, name='pixelshufflerx2/1')
n = Conv2d(n, 256, (3, 3), (1, 1), act=tf.nn.relu, padding='SAME', W_init=w_init, name='n256s1/2')
# n = SubpixelConv2d(n, scale=1, n_out_channel=None, act=tf.nn.relu, name='pixelshufflerx2/2')
# n = Conv2d(n, n_out, (1, 1), (1, 1), act=tf.nn.tanh, padding='SAME', W_init=w_init, name='out')
n = Conv2d(n, n_out, (1, 1), (1, 1), act=tf.nn.tanh, padding='SAME', W_init=w_init, name='out')
n = tf.add(n.outputs, Input.outputs)
return n
def SRresnet_no_batch(t_image, is_train=False, reuse=False):
w_init = tf.random_normal_initializer(stddev=0.02)
b_init = None # tf.constant_initializer(value=0.0)
g_init = tf.random_normal_initializer(1., 0.02)
with tf.variable_scope("SRGAN_g", reuse=reuse) as vs:
tl.layers.set_name_reuse(reuse)
n = InputLayer(t_image, name='in')
n = Conv2d(n, 64, (3, 3), (1, 1), act=tf.nn.relu, padding='SAME', W_init=w_init, name='n64s1/c')
temp = n
# B residual blocks
for i in range(16):
nn = Conv2d(n, 64, (3, 3), (1, 1), act=tf.nn.relu, padding='SAME', W_init=w_init, b_init=b_init,
name='n64s1/c1/%s' % i)
# nn = BatchNormLayer(nn, act=tf.nn.relu, is_train=is_train, gamma_init=g_init, name='n64s1/b1/%s' % i)
nn = Conv2d(nn, 64, (3, 3), (1, 1), act=None, padding='SAME', W_init=w_init, b_init=b_init,
name='n64s1/c2/%s' % i)
# nn = BatchNormLayer(nn, is_train=is_train, gamma_init=g_init, name='n64s1/b2/%s' % i)
nn = ElementwiseLayer([n, nn], tf.add, 'b_residual_add/%s' % i)
n = nn
n = Conv2d(n, 64, (3, 3), (1, 1), act=None, padding='SAME', W_init=w_init, b_init=b_init, name='n64s1/c/m')
# n = BatchNormLayer(n, is_train=is_train, gamma_init=g_init, name='n64s1/b/m')
n = ElementwiseLayer([n, temp], tf.add, 'add3')
# B residual blacks end
n = Conv2d(n, 256, (3, 3), (1, 1), act=tf.nn.relu, padding='SAME', W_init=w_init, name='n256s1/1')
n = SubpixelConv2d(n, scale=2, n_out_channel=None, act=tf.nn.relu, name='pixelshufflerx2/1')
# n = Conv2d(n, 256, (3, 3), (1, 1), act=tf.nn.relu, padding='SAME', W_init=w_init, name='n256s1/2')
# n = SubpixelConv2d(n, scale=2, n_out_channel=None, act=tf.nn.relu, name='pixelshufflerx2/2')
n = Conv2d(n, NUM_CHENNELS, (1, 1), (1, 1), act=None, padding='SAME', W_init=w_init, name='out')
tf.summary.image('final', tf.cast(n.outputs,tf.uint8), 1)
return n.outputs
def vgg7(patches, name='vgg7'):
"""
模型的输出
:param patches: input patches to improve resolution. must has format of
[batch_size, patch_height, patch_width, patch_chennels]
:param name: the name of the network
:return: the RSCNN inference function
"""
with tf.variable_scope(name):
#upscaled_patches = tf.image.resize_bicubic(patches, [INPUT_SIZE, INPUT_SIZE], True)
conv1 = conv2d(patches, 3, 3, 32, padding='SAME', name='conv1')
lrelu1 = leaky_relu(conv1, name='leaky_relu1')
conv2 = conv2d(lrelu1, 3, 3, 32, padding='SAME', name='conv2')
lrelu2 = leaky_relu(conv2, name='leaky_relu2')
conv3 = conv2d(lrelu2, 3, 3, 64, padding='SAME', name='conv3')
lrelu3 = leaky_relu(conv3, name='leaky_relu3')
conv4 = conv2d(lrelu3, 3, 3, 64, padding='SAME', name='conv4')
lrelu4 = leaky_relu(conv4, name='leaky_relu4')
conv5 = conv2d(lrelu4, 3, 3, 128, padding='SAME', name='conv5')
lrelu5 = leaky_relu(conv5, name='leaky_relu5')
conv6 = conv2d(lrelu5, 3, 3, 128, padding='SAME', name='conv6')
lrelu6 = leaky_relu(conv6, name='leaky_relu6')
return conv2d(lrelu6, 3, 3, NUM_CHENNELS, padding='SAME', name='conv_out')
def vgg_deconv_7(patches, name='vgg_deconv_7'):
with tf.variable_scope(name):
conv1 = conv2d(patches, 3, 3, 16, padding='SAME', name='conv1')
lrelu1 = leaky_relu(conv1, name='leaky_relu1')
conv2 = conv2d(lrelu1, 3, 3, 32, padding='SAME', name='conv2')
lrelu2 = leaky_relu(conv2, name='leaky_relu2')
conv3 = conv2d(lrelu2, 3, 3, 64, padding='SAME', name='conv3')
lrelu3 = leaky_relu(conv3, name='leaky_relu3')
conv4 = conv2d(lrelu3, 3, 3, 128, padding='SAME', name='conv4')
lrelu4 = leaky_relu(conv4, name='leaky_relu4')
conv5 = conv2d(lrelu4, 3, 3, 128, padding='SAME', name='conv5')
lrelu5 = leaky_relu(conv5, name='leaky_relu5')
conv6 = conv2d(lrelu5, 3, 3, 256, padding='SAME', name='conv6')
lrelu6 = leaky_relu(conv6, name='leaky_relu6')
batch_size = int(lrelu6.get_shape()[0])
rows = int(lrelu6.get_shape()[1])
cols = int(lrelu6.get_shape()[2])
channels = int(patches.get_shape()[3])
# to avoid chessboard artifacts, the filter size must be dividable by the stride
return deconv2d(lrelu6, 4, 4, [batch_size, rows, cols, channels], stride=(1, 1), name='deconv_out')
def EDSR(patches,feature_size=64,num_layers=16):
print("Building EDSR...")
# mean_x = tf.reduce_mean(patches)
# image_input = patches - mean_x
# mean_y = tf.reduce_mean(high_patches)
# image_target = y - mean_y
x = slim.conv2d(patches, feature_size, [3, 3])
conv_1 = x
"""
This creates `num_layers` number of resBlocks
a resBlock is defined in the paper as
(excuse the ugly ASCII graph)
x
|\
| \
| conv2d
| relu
| conv2d
| /
|/
+ (addition here)
|
result
"""
"""
Doing scaling here as mentioned in the paper:
`we found that increasing the number of feature
maps above a certain level would make the training procedure
numerically unstable. A similar phenomenon was
reported by Szegedy et al. We resolve this issue by
adopting the residual scaling with factor 0.1. In each
residual block, constant scaling layers are placed after the
last convolution layers. These modules stabilize the training
procedure greatly when using a large number of filters.
In the test phase, this layer can be integrated into the previous
convolution layer for the computational efficiency.'
"""
scaling_factor = 1
# Add the residual blocks to the model
for i in range(num_layers):
x = resBlock(x, feature_size, scale=scaling_factor)
# One more convolution, and then we add the output of our first conv layer
x = slim.conv2d(x, feature_size, [3, 3])
x += conv_1
# Upsample output of the convolution
x = upsample(x, NUM_CHENNELS, Scale,feature_size, None)
output =x
mean_x = tf.reduce_mean(patches)
f = tf.sqrt(tf.reduce_sum(tf.square(output + mean_x), axis=-1))
ff = tf.reshape(f, [-1, 384, 384, 1])
tf.summary.image("output_image", tf.cast(ff, tf.uint8))
# f = tf.sqrt(tf.reduce_sum(tf.square(output[:,:,:,0:2] + mean_x), axis=-1))
# ff = tf.reshape(f, [-1, 384, 384, 1])
# tf.summary.image("output_image_echo1_FA1", tf.cast(ff, tf.uint8))
# f = tf.sqrt(tf.reduce_sum(tf.square(output[:,:,:,2:4] + mean_x), axis=-1))
# ff = tf.reshape(f, [-1, 384, 384, 1])
# tf.summary.image("output_image_echo2_FA1", tf.cast(ff, tf.uint8))
# f = tf.sqrt(tf.reduce_sum(tf.square(output[:,:,:,4:6] + mean_x), axis=-1))
# ff = tf.reshape(f, [-1, 384, 384, 1])
# tf.summary.image("output_image_echo3_FA1", tf.cast(ff, tf.uint8))
# f = tf.sqrt(tf.reduce_sum(tf.square(output[:,:,:,6:8] + mean_x), axis=-1))
# ff = tf.reshape(f, [-1, 384, 384, 1])
# tf.summary.image("output_image_echo1_FA2", tf.cast(ff, tf.uint8))
# f = tf.sqrt(tf.reduce_sum(tf.square(output[:,:,:,8:10] + mean_x), axis=-1))
# ff = tf.reshape(f, [-1, 384, 384, 1])
# tf.summary.image("output_image_echo2_FA2", tf.cast(ff, tf.uint8))
# f = tf.sqrt(tf.reduce_sum(tf.square(output[:,:,:,10:12] + mean_x), axis=-1))
# ff = tf.reshape(f, [-1, 384, 384, 1])
# tf.summary.image("output_image_echo3_FA2", tf.cast(ff, tf.uint8))
return output
def loss(inferences, ground_truthes, huber_width=0.1, weights_decay=0, name='loss'):
with tf.name_scope(name):
# delta *= [[[[0.11448, 0.58661, 0.29891]]]] # weights of B, G and R
# delta *= [1]
# l2_loss = tf.pow(delta, 2)
# mse_loss = tf.reduce_mean(tf.reduce_sum(l2_loss, axis=[1, 2, 3]))
# mse_loss = tf.reduce_mean(tf.reduce_sum(l2_loss))
mse_loss = tf.reduce_mean(tf.square(inferences - ground_truthes))
return mse_loss
def loss_l1(inferences, ground_truthes, huber_width=0.1, weights_decay=0, name='loss'):
with tf.name_scope(name):
# mean_y = tf.reduce_mean(ground_truthes)
# image_target = ground_truthes - mean_y
# delta *= [[[[0.11448, 0.58661, 0.29891]]]] # weights of B, G and R
# delta *= [1]
# l2_loss = tf.pow(delta, 2)
# mse_loss = tf.reduce_mean(tf.reduce_sum(l2_loss, axis=[1, 2, 3]))
# mse_loss = tf.reduce_mean(tf.reduce_sum(l2_loss))
mse_loss = tf.reduce_mean(tf.losses.absolute_difference(ground_truthes, inferences))
return mse_loss
def PSNRLoss(inferences,ground_truthes, name = 'accurancy'):
"""
PSNR is Peek Signal to Noise Ratio, which is similar to mean squared error.
It can be calculated as
PSNR = 20 * log10(MAXp) - 10 * log10(MSE)
When providing an unscaled input, MAXp = 255. Therefore 20 * log10(255)== 48.1308036087.
However, since we are scaling our input, MAXp = 1. Therefore 20 * log10(1) = 0.
Thus we remove that component completely and only compute the remaining MSE component.
"""
with tf.name_scope(name):
accurancy = tf.reduce_mean(-10. * tf.log(tf.square(inferences-ground_truthes)))
return accurancy
# def ssim_loss(inferences, ground_truthes, BATCH_SIZE, PATCH_SIZE, Number_CHANNELS = NUM_CHENNELS, name = 'ssim_loss'):
# img1, img2 = Norm_to_gray_scale(inferences,ground_truthes,BATCH_SIZE,PATCH_SIZE,Number_CHANNELS )
# with tf.name_scope(name):
# ssim_value = tf_ssim(img1,img2)
# ssim_value2 = loss_DSSIS_tf11(img1,img2)
# return ssim_value, ssim_value2
举报
请认真填写举报原因，尽可能描述详细。
请选择举报类型
误判申诉
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手
尝试更多
代码解读
代码找茬
代码优化