I am trying to implement a paper on Image tempering detection and localization, the paper is Image Manipulation Detection and Localization Based on the Dual-Domain Convolutional Neural Networks, I was able to implement the SCNN, the one surrounded by red dots, I could not quite understand the FCNN, the one that is surrounded with blue dots.

The problem I am facing is: How the network made features vector from (1048 x 100) to (523 x 100) through max-pooling (instead of 524 x 100), and from (523 x 100) to (260 x 100) and then (260 x 100) to (256, ).

enter image description here

It appears that the given network diagram might be wrong, but, if it is wrong, how could it be published in IEEE. Please, help me understand how the FCNN is constructed.

  • $\begingroup$ How did you get 520x100? With a max pooling filter of 2x1 I would assume the output should be 524x100 $\endgroup$
    – Recessive
    Jul 20 '20 at 4:26
  • $\begingroup$ @Recessive sorry, typo it is 524 x 100 $\endgroup$
    – kiran
    Jul 20 '20 at 4:34
  • $\begingroup$ Ok that makes sense, at least the clarifies. Unfortunately I don't know why their numbers are off by one. I can only assume that they shifted the max pool filter by 1 at some point, but I can't know for sure or why. Hopefully someone else does $\endgroup$
    – Recessive
    Jul 20 '20 at 4:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.