For this week's homework, how many "Conv2D()" layers are used in the model defined in https://cross-entropy.net/ML530/imagenet-transfer.py.txt?

There are 75 Conv2D() layers in the convolutional base of the EfficientNetV2B0 model.


The following code can be used to count layers by type.  Note that you'll need a reference to "model" in order to run this code.

count = {}
for layer in model.layers:
    key = layer.__class__.__name__
    if (key in count):
        count[key] += 1
    else:
        count[key] = 1

for k,v in sorted(count.items(), key = lambda x: x[1], reverse = True):
    print(v, k, sep = '\t')


Here are the frequencies by layer type for the entire model ...

75      Conv2D
59      BatchNormalization
39      Activation
17      Dropout
17      GlobalAveragePooling2D
16      DepthwiseConv2D
16      Reshape
16      Multiply
15      Add
2       Dense
1       InputLayer
1       Resizing
1       RandomCrop
1       RandomFlip
1       Rescaling
1       Normalization