Hi everyone,
I have a simple question about the visualizing of CNN, at the time: 1:17 Jeremy speak about how the layer1 looks like, and I didn’t understand how the image was created, why this picture came from the first layer, and if someone in Imagenet builds this layer to be like that.
Thanks for the help,
Morag
Hi @moragb,
Jeremy does not go into the details of how these images are created because they are highly technical. However, I will try to explain you the gist of it. The paper Jeremy is referring to is Visualizing and Understanding Convolutional Networks.
The idea is to create a convolutional neural network with an attached deconvolutional neural network. To quote the authors, the images you are referring to are created as:
To examine a convnet, a deconvnet is attached to each
of its layers, as illustrated in Fig. 1(top), providing a
continuous path back to image pixels. To start, an
input image is presented to the convnet and features
computed throughout the layers. To examine a given
convnet activation, we set all other activations in the
layer to zero and pass the feature maps as input to
the attached deconvnet layer. Then we successively
(i) unpool, (ii) rectify and (iii) filter to reconstruct
the activity in the layer beneath that gave rise to the
chosen activation. This is then repeated until input
pixel space is reached.
If you want to know more about how the unpooling, rectifying and filtering operations work I recommend to read the paper. It is clearly written!
Best, Mark