修正input_img_data = np.random.random((1, 150, 150, 3)) * 20 + 128.

原始碼如下:

# 灰色圖像 帶有一些噪音 input_img_data = np.random.random((1, 150, 150, 3)) * 20 + 128. # 40 steps forgradient ascent step = 1. # 每個梯度更新的幅度 for i in range(40): # 計算損失值和梯度值 loss_value, grads_value = iterate([input_img_data]) # 調整輸入圖像的方向，使損失最大化 input_img_data += grads_value * step

出現錯誤:

NameError Traceback (most recent call last) Cell In[30], line 9 6 step = 1. # 每個梯度更新的幅度 7 for i in range(40): 8 # 計算損失值和梯度值 ----> 9 loss_value, grads_value = iterate([input_img_data]) 10 # 調整輸入圖像的方向，使損失最大化 11 input_img_data += grads_value * step  

NameError: name 'iterate' is not defined 

修正成這樣就可以了

import numpy as np

import tensorflow as tf

from tensorflow.keras import optimizers

# The rest of your code remains the same

# Compile your model before using it in the gradient computation

model.compile(loss='binary_crossentropy',

              optimizer=optimizers.RMSprop(learning_rate=1e-5),

              metrics=['acc'])

# Create a custom function to compute loss and gradients

@tf.function

def compute_loss_and_gradients(inputs):

    with tf.GradientTape() as tape:

        predictions = model(inputs)

        loss = tf.reduce_mean(predictions)

    gradients = tape.gradient(loss, inputs)

    return loss, gradients

# Dummy input for example gradient computation (replace with actual input)

input_height = 150  # Replace with the height of your input images

input_width = 150   # Replace with the width of your input images

input_channels = 3  # Replace with the number of channels in your input images

input_img_data = np.random.random((1, input_height, input_width, input_channels)) * 20 + 128.

# 40 steps for gradient ascent

step = 1.  # Each gradient update magnitude

for i in range(40):

    # Compute loss value and gradients

    loss_value, grads_value = compute_loss_and_gradients(tf.convert_to_tensor(input_img_data))

    if grads_value is not None:

        # Adjust the input image's direction to maximize the loss

        input_img_data += grads_value * step

    else:

        # If gradients are None, break the loop to avoid further processing

        print("Gradients are None. Stopping gradient ascent.")

        break