0
$\begingroup$

Is Siamese network rotation invariant which means if I train my siamese network on the different rotated versions of the same image so will it treat each image as different image or same.

Also if I pass for example a image of cat and a rotated image of same cat as a pair to siamese network will it give a similarity score of 0 or 1.

$\endgroup$
0
$\begingroup$

That depends on the latent representation the model has learned. Typically it won't be rotational invariant because there are no inbuilt mechanism in vanilla CNN or constraints enforcing that!

If you want to learn rotational invariant representation then either you can use a Rotation Invariant CNN or you may set up the training objective of the Siamese network such that when you pass random rotated versions of the same image of the same input, you classify as positive or else negative. The 2nd method is enforcing the constraint that your representation must be rotational invariant.

$\endgroup$
2
  • $\begingroup$ @Rishabh Rahatgaonkar Do you have any doubts? $\endgroup$ – user1825567 May 19 '20 at 20:42
  • $\begingroup$ Thank you very much $\endgroup$ – Rishabh Rahatgaonkar Sep 3 '20 at 11:10
0
$\begingroup$

Siamese network with pure CNN not rotation or scale invariant.

Found this this paper "INVARIANT DESCRIPTOR LEARNING USING A SIAMESE CONVOLUTIONAL NEURAL NETWORK"

enter image description here

$\endgroup$
1

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.