问题描述
我正在使用Python 3.7.7。和Tensorflow 2.1.0。
我有一个经过培训的U-Net网络，我想获得其编码器和其解码器。
在下图中：
您可以看到卷积编码器-解码器体系结构。我要获取编码器部分，即出现在图像左侧的层：
解码器部分：
def get_unet_uncompiled(img_shape = (200,200,1)):
inputs = Input(shape=img_shape)
conv1 = Conv2D(64,(5,5),activation='relu',padding='same',data_format="channels_last",name='conv1_1')(inputs)
conv1 = Conv2D(64,name='conv1_2')(conv1)
pool1 = MaxPooling2D(pool_size=(2,2),name='pool1')(conv1)
conv2 = Conv2D(96,(3,3),name='conv2_1')(pool1)
conv2 = Conv2D(96,name='conv2_2')(conv2)
pool2 = MaxPooling2D(pool_size=(2,name='pool2')(conv2)
conv3 = Conv2D(128,name='conv3_1')(pool2)
conv3 = Conv2D(128,name='conv3_2')(conv3)
pool3 = MaxPooling2D(pool_size=(2,name='pool3')(conv3)
conv4 = Conv2D(256,name='conv4_1')(pool3)
conv4 = Conv2D(256,(4,4),name='conv4_2')(conv4)
pool4 = MaxPooling2D(pool_size=(2,name='pool4')(conv4)
conv5 = Conv2D(512,name='conv5_1')(pool4)
conv5 = Conv2D(512,name='conv5_2')(conv5)
up_conv5 = UpSampling2D(size=(2,name='up_conv5')(conv5)
ch,cw = get_crop_shape(conv4,up_conv5)
crop_conv4 = Cropping2D(cropping=(ch,cw),name='crop_conv4')(conv4)
up6 = concatenate([up_conv5,crop_conv4])
conv6 = Conv2D(256,name='conv6_1')(up6)
conv6 = Conv2D(256,name='conv6_2')(conv6)
up_conv6 = UpSampling2D(size=(2,name='up_conv6')(conv6)
ch,cw = get_crop_shape(conv3,up_conv6)
crop_conv3 = Cropping2D(cropping=(ch,name='crop_conv3')(conv3)
up7 = concatenate([up_conv6,crop_conv3])
conv7 = Conv2D(128,name='conv7_1')(up7)
conv7 = Conv2D(128,name='conv7_2')(conv7)
up_conv7 = UpSampling2D(size=(2,name='up_conv7')(conv7)
ch,cw = get_crop_shape(conv2,up_conv7)
crop_conv2 = Cropping2D(cropping=(ch,name='crop_conv2')(conv2)
up8 = concatenate([up_conv7,crop_conv2])
conv8 = Conv2D(96,name='conv8_1')(up8)
conv8 = Conv2D(96,name='conv8_2')(conv8)
up_conv8 = UpSampling2D(size=(2,name='up_conv8')(conv8)
ch,cw = get_crop_shape(conv1,up_conv8)
crop_conv1 = Cropping2D(cropping=(ch,name='crop_conv1')(conv1)
up9 = concatenate([up_conv8,crop_conv1])
conv9 = Conv2D(64,name='conv9_1')(up9)
conv9 = Conv2D(64,name='conv9_2')(conv9)
ch,cw = get_crop_shape(inputs,conv9)
conv9 = ZeroPadding2D(padding=(ch,name='conv9_3')(conv9)
conv10 = Conv2D(1,(1,1),activation='sigmoid',name='conv10_1')(conv9)
model = Model(inputs=inputs,outputs=conv10)
return model
辅助功能是：
def get_crop_shape(target,refer):
# width,the 3rd dimension
cw = (target.get_shape()[2] - refer.get_shape()[2])
assert (cw >= 0)
if cw % 2 != 0:
cw1,cw2 = cw // 2,cw // 2 + 1
else:
cw1,cw // 2
# height,the 2nd dimension
ch = (target.get_shape()[1] - refer.get_shape()[1])
assert (ch >= 0)
if ch % 2 != 0:
ch1,ch2 = ch // 2,ch // 2 + 1
else:
ch1,ch // 2
return (ch1,ch2),(cw1,cw2)
此模型的图形为：
我训练网络，训练后我得到编码器执行此操作：
first_encoder_layer = 0
last_encoder_layer = 14
old_model = get_unet_uncompiled()
old_model.compile(tf.keras.optimizers.Adam(lr=(1e-4) * 2),loss='binary_crossentropy',metrics=['accuracy'])
encoder: Model = Model(inputs=old_model.layers[first_encoder_layer].input,outputs=old_model.layers[last_encoder_layer].output,name='encoder')
它运行完美。
decoder: Model = Model(inputs=old_model.layers[last_encoder_layer + 1].input,outputs=old_model.layers[-1].output,name='decoder')
我得到了我不理解的错误：
WARNING:tensorflow:Model inputs must come from `tf.keras.Input` (thus holding past layer Metadata),they cannot be the output of a prevIoUs non-Input layer. Here,a tensor specified as input to "decoder" was not an Input tensor,it was generated by layer up_conv5.
Note that input tensors are instantiated via `tensor = tf.keras.Input(shape)`.
The tensor that caused the issue was: up_conv5/Identity:0
WARNING:tensorflow:Model inputs must come from `tf.keras.Input` (thus holding past layer Metadata),it was generated by layer crop_conv4.
Note that input tensors are instantiated via `tensor = tf.keras.Input(shape)`.
The tensor that caused the issue was: crop_conv4/Identity:0
ValueError: Graph disconnected: cannot obtain value for tensor Tensor("input_1:0",shape=(None,dtype=float32) at layer "input_1". The following prevIoUs layers were accessed without issue: []
decoder_input = Input(shape=(12,12,512),name='dec_input')
z = UpSampling2D(size=(2,name='up_dec_conv5')(decoder_input)
decoder: Model = Model(inputs=z,name='decoder')
但是我几乎得到了相同的错误：
WARNING:tensorflow:Model inputs must come from `tf.keras.Input` (thus holding past layer Metadata),it was generated by layer up_dec_conv5.
Note that input tensors are instantiated via `tensor = tf.keras.Input(shape)`.
The tensor that caused the issue was: up_dec_conv5/Identity:0
如何从预训练的U-Net网络获得解码器？
解决方法
我建议为编码器和解码器（get_encoder，get_decoder）定义一个结构。在训练完整个模型之后，我们的想法是创建一个新的解码器架构（通过get_decoder），我们可以用解码器训练后的权重填充
pythonic语言，您可以通过这种方式实现...
def get_crop_shape(target,refer):
# width,the 3rd dimension
cw = (target.get_shape()[2] - refer.get_shape()[2])
assert (cw >= 0)
if cw % 2 != 0:
cw1,cw2 = cw // 2,cw // 2 + 1
else:
cw1,cw // 2
# height,the 2nd dimension
ch = (target.get_shape()[1] - refer.get_shape()[1])
assert (ch >= 0)
if ch % 2 != 0:
ch1,ch2 = ch // 2,ch // 2 + 1
else:
ch1,ch // 2
return (ch1,ch2),(cw1,cw2)
def get_encoder(img_shape):
inp = Input(shape=img_shape)
conv1 = Conv2D(64,(5,5),activation='relu',padding='same',data_format="channels_last",name='conv1_1')(inp)
conv1 = Conv2D(64,name='conv1_2')(conv1)
pool1 = MaxPooling2D(pool_size=(2,2),name='pool1')(conv1)
conv2 = Conv2D(96,(3,3),name='conv2_1')(pool1)
conv2 = Conv2D(96,name='conv2_2')(conv2)
pool2 = MaxPooling2D(pool_size=(2,name='pool2')(conv2)
conv3 = Conv2D(128,name='conv3_1')(pool2)
conv3 = Conv2D(128,name='conv3_2')(conv3)
pool3 = MaxPooling2D(pool_size=(2,name='pool3')(conv3)
conv4 = Conv2D(256,name='conv4_1')(pool3)
conv4 = Conv2D(256,(4,4),name='conv4_2')(conv4)
pool4 = MaxPooling2D(pool_size=(2,name='pool4')(conv4)
conv5 = Conv2D(512,name='conv5_1')(pool4)
conv5 = Conv2D(512,name='conv5_2')(conv5)
return conv5,conv4,conv3,conv2,conv1,inp
def get_decoder(convs):
conv5,inputs = convs
up_conv5 = UpSampling2D(size=(2,name='up_conv5')(conv5)
ch,cw = get_crop_shape(conv4,up_conv5)
crop_conv4 = Cropping2D(cropping=(ch,cw),name='crop_conv4')(conv4)
up6 = concatenate([up_conv5,crop_conv4])
conv6 = Conv2D(256,name='conv6_1')(up6)
conv6 = Conv2D(256,name='conv6_2')(conv6)
up_conv6 = UpSampling2D(size=(2,name='up_conv6')(conv6)
ch,cw = get_crop_shape(conv3,up_conv6)
crop_conv3 = Cropping2D(cropping=(ch,name='crop_conv3')(conv3)
up7 = concatenate([up_conv6,crop_conv3])
conv7 = Conv2D(128,name='conv7_1')(up7)
conv7 = Conv2D(128,name='conv7_2')(conv7)
up_conv7 = UpSampling2D(size=(2,name='up_conv7')(conv7)
ch,cw = get_crop_shape(conv2,up_conv7)
crop_conv2 = Cropping2D(cropping=(ch,name='crop_conv2')(conv2)
up8 = concatenate([up_conv7,crop_conv2])
conv8 = Conv2D(96,name='conv8_1')(up8)
conv8 = Conv2D(96,name='conv8_2')(conv8)
up_conv8 = UpSampling2D(size=(2,name='up_conv8')(conv8)
ch,cw = get_crop_shape(conv1,up_conv8)
crop_conv1 = Cropping2D(cropping=(ch,name='crop_conv1')(conv1)
up9 = concatenate([up_conv8,crop_conv1])
conv9 = Conv2D(64,name='conv9_1')(up9)
conv9 = Conv2D(64,name='conv9_2')(conv9)
ch,cw = get_crop_shape(inputs,conv9)
conv9 = ZeroPadding2D(padding=(ch,name='conv9_3')(conv9)
conv10 = Conv2D(1,(1,1),activation='sigmoid',name='conv10_1')(conv9)
return conv10
def get_unet(img_shape = (200,200,1)):
enc = get_encoder(img_shape)
dec = get_decoder(enc)
model = Model(inputs=enc[-1],outputs=dec)
return model
创建整个模型并拟合
img_shape = (200,1)
old_model = get_unet(img_shape)
# old_model.compile(...)
# old_model.fit(...)
一如既往地提取编码器
# extract encoder
first_encoder_layer = 0
last_encoder_layer = 14
encoder_output_layer = [14,11,8,5,2,0]
encoder = Model(inputs=old_model.layers[first_encoder_layer].input,outputs=[old_model.layers[l].output for l in encoder_output_layer],name='encoder')
encoder.summary()
创建解码器结构并分配训练后的权重
# extract decoder fitted weights
restored_w = []
for w in old_model.layers[last_encoder_layer + 1:]:
restored_w.extend(w.get_weights())
# reconstruct decoder architecture setting the fitted weights
new_inp = [Input(l.shape[1:]) for l in get_encoder(img_shape)]
new_dec = get_decoder(new_inp)
decoder = Model(new_inp,new_dec)
decoder.set_weights(restored_w)
decoder.summary()
收益预测
# generate random images
n_images = 20
X = np.random.uniform(0,1,(n_images,1)).astype('float32')
# get encoder predictions
pred_encoder = encoder.predict(X)
print([p.shape for p in pred_encoder])
# get decoder predictions
pred_decoder = decoder.predict(pred_encoder)
print(pred_decoder.shape)
here正在运行的笔记本