Grad-CAM - Which layer to choose?

[Danielrcnn]

trafficstars

Hi, which layer would you choose to be the last convolution layer?

My network model is:

_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 mobilenetv2_1.00_224 (Funct  (None, 7, 7, 1280)       2257984   
 ional)                                                          
                                                                 
 flatten_2 (Flatten)         (None, 62720)             0         
                                                                 
 dense_8 (Dense)             (None, 64)                4014144   
                                                                 
 batch_normalization_6 (Batc  (None, 64)               256       
 hNormalization)                                                 
                                                                 
 activation_6 (Activation)   (None, 64)                0         
                                                                 
 dropout_6 (Dropout)         (None, 64)                0         
                                                                 
 dense_9 (Dense)             (None, 32)                2080      
                                                                 
 batch_normalization_7 (Batc  (None, 32)               128       
 hNormalization)                                                 
                                                                 
 activation_7 (Activation)   (None, 32)                0         
                                                                 
 dropout_7 (Dropout)         (None, 32)                0         
                                                                 
 dense_10 (Dense)            (None, 2)                 66        
                                                                 
=================================================================
Total params: 6,274,658
Trainable params: 4,016,482
Non-trainable params: 2,258,176
_________________________________________________________________

When choosing dense_10 the following error appears:

InvalidArgumentError                      Traceback (most recent call last)
g:\Outros computadores\Meu modelo Computador\IFES\10 Período\TCC\Códigos\GradCAM.ipynb Célula: 5 in <cell line: 16>()
     13 print(preds)
     15 # Generate class activation heatmap
---> 16 heatmap = make_gradcam_heatmap(img_array, model, last_conv_layer_name)
     18 # Display heatmap
     19 plt.matshow(heatmap)

g:\Outros computadores\Meu modelo Computador\IFES\10 Período\TCC\Códigos\GradCAM.ipynb Célula: 5 in make_gradcam_heatmap(img_array, model, last_conv_layer_name, pred_index)
     29 grads = tape.gradient(class_channel, last_conv_layer_output)
     31 # This is a vector where each entry is the mean intensity of the gradient
     32 # over a specific feature map channel
---> 33 pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))
     35 # We multiply each channel in the feature map array
     36 # by "how important this channel is" with regard to the top predicted class
     37 # then sum all the channels to obtain the heatmap class activation
     38 last_conv_layer_output = last_conv_layer_output[0]

File c:\Users\danie\AppData\Local\Programs\Python\Python310\lib\site-packages\tensorflow\python\util\traceback_utils.py:153, in filter_traceback.<locals>.error_handler(*args, **kwargs)
    151 except Exception as e:
    152   filtered_tb = _process_traceback_frames(e.__traceback__)
--> 153   raise e.with_traceback(filtered_tb) from None
    154 finally:
    155   del filtered_tb
...
     55                                       inputs, attrs, num_outputs)
     56 except core._NotOkStatusException as e:
     57   if name is not None:

InvalidArgumentError: Invalid reduction dimension (2 for input with 2 dimension(s) [Op:Mean]

How can I resolve this?