0
$\begingroup$

I'm currently working on a common object classification and I use a CNN to do that. Each instance has 3 images with differents views (front, back, side) and I would like to use this to improve my network, so I want to build a multiple input network. I'm wondering if I should :

  • Build 1 network with 3 different pipeline convolutional, merge their result and then use a softmax.

  • Build 1 network with shared weigth for features part and then softmax

$\endgroup$
1
$\begingroup$

The best way to find out the answer to your question is to simply try out both approaches and see which works better.

This may sound stupid or obvious, but the architecture of the network is simply another "hyperparameter" which you need to tune and every dataset may need something else.

That said, I would personally bet on a combination of these: In the first half of the net, share the weights between three pipelines, then concatenate them and apply several additional conv. layers before the final softmax.

$\endgroup$
  • $\begingroup$ Hi, thanks for your answer, I don't understand quite 'share the weights between three pipelines', because either i have 3 pipeline, either i share 1 pipeline for 3 channels. $\endgroup$ – alexandre_d Mar 16 '18 at 11:15

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.