我正在训练一个基于多标签VGG-16的分类模型。这个任务有25个标签。我试图在https://towardsdatascience.com/multi-label-classification-and-class-activation-map-on-fashion-mnist-1454f09f5925复制这个代码，使用训练好的模型生成类激活图。

model = load_model('weights/vgg16_multilabel.09-0.3833.h5')    
model.summary()    
sgd = SGD(learning_rate=0.001, decay=1e-6, momentum=0.9, nesterov=True)     
model.compile(optimizer=sgd, loss='binary_crossentropy',     
                     metrics=['accuracy'])  

#labels
columns=['Action', 'Adventure', 'Animation', 'Biography', 'Comedy',     
         'Crime', 'Documentary', 'Drama', 'Family', 'Fantasy',    
         'History', 'Horror', 'Music', 'Musical', 'Mystery',    
         'N/A', 'News', 'Reality-TV', 'Romance', 'Sci-Fi', 'Short',    
         'Sport', 'Thriller', 'War', 'Western']      

gap_weights = model.layers[-1].get_weights()[0] #final dense layer    
print(" >>> size(gap_weights) = ", gap_weights.size)  

#extract from the deepest convolutional layer

cam_model = Model(inputs=model.input,     
                    outputs=(model.layers[-3].output,     
                             model.layers[-1].output)) 

print(" >>> K.int_shape(model.layers[-3].output) = ", K.int_shape(model.layers[-3].output))    
print(" >>> K.int_shape(model.layers[-1].output) = ", K.int_shape(model.layers[-1].output))   

#--- make the prediction    
features, results = cam_model.predict(X_test)   

# check the CAM activations for 10 test images    
for idx in range(10):      
    # get the feature map of the test image    
    features_for_one_img = features[idx, :, :, :]       

    # map the feature map to the original size    
    height_roomout = train_img_size_h / features_for_one_img.shape[0]    
    width_roomout = train_img_size_w / features_for_one_img.shape[1]    
    cam_features = sp.ndimage.zoom(features_for_one_img, (height_roomout, width_roomout, 1), order=2)       
    # get the predicted label with the maximum probability    
    pred = np.argmax(results[idx])   
    
    # prepare the final display    
    plt.figure(facecolor='white')

    # get the weights of class activation map    
    cam_weights = gap_weights[:, pred]

    # create the class activation map    
    cam_output = np.dot(cam_features, cam_weights)

    
    # draw the class activation map    
    ax.set_xticklabels([])    
    ax.set_yticklabels([])    
    
    buf = 'Predicted Class = ' + columns[pred] + ', Probability = ' + str(results[idx][pred])

    plt.xlabel(buf)    
    plt.imshow(t_pic[idx], alpha=0.5)    
    plt.imshow(cam_output, cmap='jet', alpha=0.5)  
    plt.show()

这是输出

size(gap_weights) =  12800
K.int_shape(model.layers[-4].output) =  (None, 512)
K.int_shape(model.layers[-1].output) =  (None, 25)

出现以下错误：

Traceback (most recent call last):

  File "/project/1/complete_code.py", line 1295, in <module>
    features_for_one_img = features[idx, :, :, :]

IndexError: too many indices for array: array is 2-dimensional, but 4 were indexed

我在Tensorflow 2.X中遇到这个错误，但在Tensorflow 1.X中没有遇到任何问题。