I recently asked about creating competitive images for a simple Softmax regression model. I managed to find a solution myself. Now I would like to do the same, but instead for Convnet, as described in the TensorFlow Deep MNIST tutorial for experts .
In the previous question, the situation was very simple, since the weight matrix has exactly the same dimension as the image, so we can simply add the matrix as follows:
images_fool = x + 1.5 * w_six
where images_fool is the contention image, x is the original image of the number 2, and w_six is the weight matrix for the number 6, so we can make the contention image of the number 2, which deceives the classifier, which will classify the image as the number 6.
But now, with ConvNet, there are many more weight matrices, and not one of them is the same size as the original image. So my question is: how do we add weights to the original image to create a competitive image?
I appreciate any help. Thanks!
The script can be found here .
matrix image deep-learning tensorflow conv-neural-network
George Liu
source share