DenseNet | DenseNet implementation in Keras | Machine Learning library

You can do something like this

Following this question-answer1, you should first train your model with (let's say) one input and one output. And later if you want to compute grad-cam, you would pick some intermediate layer of your base model (not the final output of the base model) and in that case, you need to build your feature extractor separately. For example

As @Frightera suggested, you are mixing keras and tensorflow.keras imports. Try the code with all tensorflow.keras imports,

Try calling the evaluate() like this:

I found you can use .get_layer() twice to acces layers inside functional densenet model embebeed in the "main" model.

In this case I can use model.get_layer('densenet121').summary() to check all thje layer inside the embebeed model, and then use them with this code: model.get_layer('densenet121').get_layer('xxxxx')

When paper introduce they method they said:

The attention modules aim to exploit the relationship between disease labels and (1) diagnosis-specific feature channels, (2) diagnosis-specific locations on images (i.e. the regions of thoracic abnormalities), and (3) diagnosis-specific scales of the feature maps.

(1), (2), (3) corresponding to channel-wise attention, element-wise attention, scale-wise attention

We can tell that element-wise attention is for deal with disease location & weight info, i.e: at each location on image, how likely there is a disease, as it been mention again when paper introduce the element-wise attention：

The element-wise attention learning aims to enhance the sensitivity of feature representations to thoracic abnormal regions, while suppressing the activations when there is no abnormality.

OK, we could easily get location & weight info for one disease, but we have multiple disease:

Since there are multiple thoracic diseases, we choose to estimate an element-wise attention map for each category in this work.

We could store the multiple disease location & weight info by using a tensor A with shape (height, width, number of disease):

The all-category attention map is denoted by A ∈ RH×W×C, where each element aijc is expected to represent the relative importance at location (i, j) for identifying the c-th category of thoracic abnormalities.

And we have linear classifiers for produce a tensor S with same shape as A, this can be interpret as:

At each location on feature maps X(CA), how confident those linear classifiers think there is certain disease at that location

Now we element-wise multiply S and A to get M, i.e we are:

prevent the attention maps from paying unnecessary attention to those location with non-existent labels

So after all those, we get tensor M which tells us:

location & weight info about certain disease that linear classifiers are confident about it

Then if we do global average pooling over M, we get prediction of weight for each disease, add another softmax (or sigmoid) we could get prediction of probability for each disease

Now since we have label and prediction, so, naturally we could minimizing loss function to optimize the model.

Following code is tested on colab and will show you how to implement channel-wise attention and element-wise attention, and build and training a simple model base on your code with DenseNet121 and without scale-wise attention:

I thought it might be much complex but it's actually rather very simple. We just need to build a model with desired output layers at the __init__ method and use it normally in the call method.

I believe the input_shape is (128, 216, 1)

The issue here is that you don't have a time-axis to time distribute your CNN (DenseNet169) layer over.

In this step -

I found out the (silly) reason behind the long training epoch:

Data consists of train_size training data and val_size validation data without considering batches. for example, training data consists of 4886 data samples which would be 76 data batches (with batch_size=64).

when I use for batch_idx, (x, y) in enumerate(train_gen):, I have a total number of 76 batches but I loop through 4886 batches in the loop by mistake.

I rewrote the following lines to these: