26、自动编码器：从卷积到超分辨率的深入探索

自动编码器：从卷积到超分辨率的深入探索

1. 自动编码器基础

自动编码器旨在学习低维表示（潜在空间），然后根据预定义任务（如恒等函数）学习一种变换来重建图像。以下是使用MNIST数据集构建和训练自动编码器的示例代码：

layers = [ {'n_nodes': 256 }, { 'n_nodes': 128 }, { 'n_nodes': 64 } ] 
inputs = Input((28, 28, 1))     
encoding = encoder(inputs, layers)
outputs = decoder(encoding, layers, (28, 28, 1))
ae = Model(inputs, outputs)
from tensorflow.keras.datasets import mnist
import numpy as np
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = (x_train / 255.0).astype(np.float32)
x_test  = (x_test  / 255.0).astype(np.float32)
x_train = np.expand_dims(x_train, axis=-1)
x_test  = np.expand_dims(x_test, axis=-1)
ae.compile(loss='binary_crossentropy', optimizer='adam', 
metrics=['accuracy'])
ae.fit(x_train, x_train, ep