tensorflow2实现多个输出

首先需要创建输出列表

model = Model(inpt, [out1, out2, out3])

然后标签也需要多个

model.fit(image, [tar1, tar2, tar3], batch_size=1)

详细代码

import tensorflow as tf
from tensorflow.keras.layers import Conv2D, GlobalAveragePooling2D, Dense
from tensorflow.keras import Model, Input
import numpy as np

#输入图像
image = np.random.normal(size=(1, 5, 5, 3))
tar1 = np.random.normal(size=(1, 3))
tar2 = np.random.normal(size=(1, 3))
tar3 = np.random.normal(size=(1, 3))

inpt = Input(shape=(5, 5, 3))
x = Conv2D(3, 2, padding='same')(inpt)

out1 = GlobalAveragePooling2D(name="out1")(x)

out2 = GlobalAveragePooling2D(name="out2")(x)

out3 = GlobalAveragePooling2D(name="out3")(x)


model = Model(inpt, [out1, out2, out3])

adam = tf.keras.optimizers.Adam(lr = 0.05, epsilon = 0.1)
model.compile(optimizer=adam, loss='mean_squared_error')

model.fit(image, [tar1, tar2, tar3], batch_size=1)

也可以自定义方法

def f(y_true, y_pred):
    result = tf.reduce_sum((y_true - y_pred)**2)
    return result

adam = tf.keras.optimizers.Adam(lr = 0.05, epsilon = 0.1)
model.compile(optimizer=adam, loss=f)

model.fit(image, [tar1, tar2, tar3], batch_size=1)

当然还有另外一种写法，使用concatenate

out = Concatenate()([out1, out2, out3])

model = Model(inpt, out)

def f(y_true, y_pred):
    result = tf.reduce_sum((y_true - y_pred)**2)
    return result

adam = tf.keras.optimizers.Adam(lr = 0.05, epsilon = 0.1)
model.compile(optimizer=adam, loss=f)

model.fit(image, np.concatenate([tar1, tar2, tar3], axis=-1), batch_size=1)