Visualizing Intermediate Representations

Hi there,

I have a question about the ‘C1_W4_Lab_2_image_generator_with_validation.ipynb’ lab. In the view of the Intermediate Representations, I am not sure if the input Neural Network receives only the last image of the last line, or if it receives the entire last line as a single image that concatenates all the images coming from the filters. Could you help me clarify this?


The plots are generated for a single random image. Please see this comment:

# Run the image through the network, thus obtaining all
# intermediate representations for this image.

That the graphics were generated for a single random image I already understood.
But what I’m asking is does the sequential images represent an input layer or does just one followed image represent an input?
Could you help me?

Hi Raphael,

Thanks for the great question,

The intermediate representation code is taking an image and highlights the feature transformation that takes place within each layer (passing an image through filters) and also through consecutive layers of the model.
When you look at the generated image after the code is run, you would see the image features transformed after passing through filters within the layer (x-axis ) and through the layers (y-axis) starting for the first layer (dense then max pooling layer) until it goes through the final layer (conv2d and maxPooling).

I hope this answers your question,

Thanks again for the question,