def test(dec_model, cls_model, red_test_dat, orange_test_dat):
# Red to Orange test
code = np.expand_dims(np.array([0., 1.]), axis=0)
for i in range(len(red_test_dat)):
if i == 0:
target_c = code
else:
target_c = np.append(target_c, code, axis=0)
c1, c2, c3 = G_utils.codemap(target_c=target_c)
tst_map = dec_model({
"dec_in": red_test_dat,
"c1": c1,
"c2": c2,
"c3": c3
},
training=False)["dec_out"]
pseudo_image = red_test_dat + tst_map
res = cls_model({"cls_in": pseudo_image}, training=False)["cls_out"]
acc = G_losses.acc(target_c, res)
stack_p = np.empty((len(red_test_dat), 1))
a = G_layers.flatten()(tst_map)
b = G_layers.flatten()(-(red_test_dat - orange_test_dat))
for i in range(len(red_test_dat)):
stack_p[i] = NCC(a[i], b[i])
# Orange to Red test
code = np.expand_dims(np.array([1., 0.]), axis=0)
for i in range(len(orange_test_dat)):
if i == 0: target_c = code
else: target_c = np.append(target_c, code, axis=0)
c1, c2, c3 = G_utils.codemap(target_c=target_c)
tst_map = dec_model(
{
"dec_in": orange_test_dat,
"c1": c1,
"c2": c2,
"c3": c3
},
training=False)["dec_out"]
pseudo_image = orange_test_dat + tst_map
res = cls_model({"cls_in": pseudo_image}, training=False)["cls_out"]
acc += G_losses.acc(target_c, res)
stack_n = np.empty((len(orange_test_dat), 1))
a = G_layers.flatten()(tst_map)
b = G_layers.flatten()(red_test_dat - orange_test_dat)
for i in range(len(orange_test_dat)):
stack_n[i] = NCC(a[i], b[i])
print("ACC: %4f | NCC(+): %4f, %4f | NCC(-): %4f, %4f" %
(acc / 2, np.mean(stack_p), np.std(stack_p), np.mean(stack_n),
np.std(stack_n)))
return acc / 2
def dense_layer(self, x, f, act="leakyrelu", n=None):
dense_l = G_layers.dense(f, act=None, name=n + "_dense")
out = dense_l(x)
if act:
act_l = G_layers.leaky_relu(n + "_leakyrelu")
out = act_l(out)
return out
def dense_layer(self, x, f, act="relu", n=None):
dense_l = G_layers.dense(f, act=None, name=n + "_dense")
out = dense_l(x)
self.layers[n+"_dense"] = dense_l
if act:
act_l = G_layers.relu(n + "_relu")
self.layers[n+"_relu"] = act_l
out = act_l(out)
return out
def conv_bn_act(self, x, f, n, s, k=None, p="same", rank=2, batch=True, act=True):
c_layer = G_layers.conv
if k:
conv_l = c_layer(f=f, k=k, p=p, s=s, rank=rank, dilation_rate=(1,1), name=n+"_conv")
else:
conv_l = c_layer(f=f, p=p, s=s, rank=rank, dilation_rate=(1,1), name=n + "_conv")
out = conv_l(x)
if batch:
batch_l = G_layers.batch_norm(name=n + "_batchnorm")
out = batch_l(out)
if act:
act_l = G_layers.leaky_relu(name=n + "_leakyrelu")
out = act_l(out)
return out
def conv_bn_act(self, x, f, n, s, k=None, p="same", rank=2, act=True, trans=False, out_p="auto"):
if trans:
c_layer = G_layers.conv_transpose
else:
c_layer = G_layers.conv
if k:
conv_l = c_layer(f=f, p=p, k=k, s=s, out_p=out_p, rank=rank, dilation_rate=(1,1), name=n+"_conv")
else:
conv_l = c_layer(f=f, p=p, rank=rank, dilation_rate=(1,1), name=n+"_conv")
out = conv_l(x)
norm_l = G_layers.batch_norm(name=n + "_norm")
out = norm_l(out)
self.layers[n+"_conv"] = conv_l
self.layers[n+"_norm"] = norm_l
if act:
act_l = G_layers.relu(name=n + "_relu")
self.layers[n + "_relu"] = act_l
out = act_l(out)
return out
def pretraining_clf(self):
# Encoder
enc_conv1 = self.conv_bn_act(x=self.enc_in_layer, k=4, s=2, f=self.ch, n="enc_conv1")
enc_conv2 = self.conv_bn_act(x=enc_conv1, k=4, s=2, f=self.ch, n="enc_conv2")
enc_conv3 = self.conv_bn_act(x=enc_conv2, k=4, s=2, f=self.ch * 2, n="enc_conv3")
enc_conv4 = self.conv_bn_act(x=enc_conv3, k=4, s=2, f=self.ch * 2, n="enc_conv4")
# Classifier
flatten = self.flatten_layer(x=enc_conv4, n="flatten")
dense1 = self.dense_layer(x=flatten, f=256, n="dense1")
dense2 = self.dense_layer(x=dense1, f=2, act=None, n="dense2")
cls_out = G_layers.softmax(dense2, name="softmax")
self.enc_model = keras.Model({"enc_in": self.enc_in_layer}, {"enc_out": enc_conv4}, name="enc_model")
self.cls_model = keras.Model({"cls_in": self.enc_in_layer}, {"cls_out": cls_out}, name="cls_model")
return self.enc_model, self.cls_model
def CFmap_generator(self):
# Encoder
enc_conv1 = self.conv_bn_act(x=self.enc_in_layer, k=4, s=2, f=self.ch, n="enc_conv1")
enc_conv2 = self.conv_bn_act(x=enc_conv1, k=4, s=2, f=self.ch, n="enc_conv2")
enc_conv3 = self.conv_bn_act(x=enc_conv2, k=4, s=2, f=self.ch*2, n="enc_conv3")
enc_conv4 = self.conv_bn_act(x=enc_conv3, k=4, s=2, f=self.ch*2, n="enc_conv4")
# Decoder
dec_up3 = G_layers.upsample(rank=2, name="dec_up3")(enc_conv4)
dec_code_concat3 = self.concat(enc_conv3, self.c3, n="dec_code_concat3")
dec_code_conv3 = self.conv_bn_act(x=dec_code_concat3, f=self.ch*2, k=3, s=1, p="same", n="dec_code_conv3")
dec_concat3 = self.concat(dec_code_conv3, dec_up3, n="dec_concat3")
dec_conv3 = self.conv_bn_act(x=dec_concat3, f=self.ch*2, k=3, s=1, p="same", n="dec_conv3")
dec_up2 = G_layers.upsample(rank=2, name="dec_up2")(dec_conv3)
dec_code_concat2 = self.concat(enc_conv2, self.c2, n="dec_code_concat2")
dec_code_conv2 = self.conv_bn_act(x=dec_code_concat2, f=self.ch, k=3, s=1, p="same", n="dec_code_conv2")
dec_concat2 = self.concat(dec_code_conv2, dec_up2, n="dec_concat2")
dec_conv2 = self.conv_bn_act(x=dec_concat2, f=self.ch, k=3, s=1, p="same", n="dec_conv2")
dec_up1 = G_layers.upsample(rank=2, name="dec_up1")(dec_conv2)
dec_code_concat1 = self.concat(enc_conv1, self.c1, n='dec_code_concat1')
dec_code_conv1 = self.conv_bn_act(x=dec_code_concat1, f=self.ch, k=3, s=1, p="same", n="dec_code_conv1")
dec_concat1 = self.concat(dec_code_conv1, dec_up1, n="dec_concat1")
dec_conv1 = self.conv_bn_act(x=dec_concat1, f=self.ch, k=3, s=1, p="same", n="dec_conv1")
dec_up = G_layers.upsample(rank=2, name="dec_up")(dec_conv1)
dec_out = self.conv_bn_act(x=dec_up, f=3, k=1, s=1, act=False, p="same", n="dec_out")
dec_out = G_layers.tanh(x=dec_out, name="dec_out_tanh")
self.dec_model = keras.Model({"dec_in": self.enc_in_layer, "c1": self.c1, "c2": self.c2, "c3": self.c3},
{"dec_out": dec_out}, name="dec_model")
# Classifier
enc_conv1 = self.conv_bn_act_reuse(x=self.enc_in_layer, n="enc_conv1")
enc_conv2 = self.conv_bn_act_reuse(x=enc_conv1, n='enc_conv2')
enc_conv3 = self.conv_bn_act_reuse(x=enc_conv2, n="enc_conv3")
enc_conv4 = self.conv_bn_act_reuse(x=enc_conv3, n='enc_conv4')
flatten = self.flatten_layer(x=enc_conv4, n="flatten")
dense1 = self.dense_layer(x=flatten, f=256, n='dense1')
dense2 = self.dense_layer(x=dense1, f=2, act=None, n="dense2")
self.cls_out = G_layers.softmax(dense2, name="softmax")
self.cls_model = keras.Model({"cls_in": self.enc_in_layer}, {"cls_out": self.cls_out}, name="cls_model")
return self.cls_model, self.dec_model
def __init__(self, ch=conf.cfmap_ch):
self.ch = ch
self.layers = {}
self.c1, self.c2, self.c3 = G_layers.geo_c1(name="c1"), G_layers.geo_c2(name="c2"), G_layers.geo_c3(name="c3")
self.enc_in_layer = G_layers.input_layer2d(name="enc_in")
self.pretraining_clf(), self.CFmap_generator()
def flatten_layer(self, x, n=None):
flatten_l = G_layers.flatten(n + "_flatten")(x)
self.layers[n+"_flatten"] = flatten_l
return flatten_l
def concat(self, x, y, n):
concat_l = G_layers.concat(name=n + "_concat")
self.layers[n+"_concat"] = concat_l
return concat_l([x, y])
def flatten_layer(self, x, n=None):
flatten_l = G_layers.flatten(n + "_flatten")(x)
return flatten_l
def __init__(self, ch=conf.disc_ch):
self.ch = ch
self.discri_in_layer = G_layers.input_layer2d(name="discriminator_in")
tf.keras.backend.set_image_data_format("channels_last")
self.build_model()