0
$\begingroup$

I want to train a CNN (Vggnet) to identify different types of buildings from aerial images.

However seeing that a CNN "ignores" size, e.g. the same type of dog in one image can be large and small in another image but will still be classified as a dog.

My issue is that non-residential buildings are mostly larger than residential houses, now I want to use this property to distinguish between residential and non residential. Is this even possible?

enter image description hereenter image description here

Thanks

$\endgroup$
  • 1
    $\begingroup$ Why don't you try and see? $\endgroup$ – DuttaA Jul 1 '18 at 7:27
  • $\begingroup$ CNN doesn't ignore size at all,who told you.Can you give us a hint! $\endgroup$ – quintumnia Jul 1 '18 at 18:32
0
$\begingroup$

You need to know the basic mechanics of CNN , you can't simply finetune a pretrained net (VGGNET) everytime. VGGnet is trained on imagenet dataset so most of the features in it are not so relevant for your task (recognising buildings from aerial view). so it is much better to train a cnn from scratch. You can't simply say that it fails to classify residential , non-residential . It is best to find what could be done by trying,

My issue is that non-residential buildings are mostly larger than residential houses, now I want to use this property to distinguish between residential and non residential.

Most of the times it is better to observe the features recognised by the net and then come with an approach , rather than enforcing your views.

You may find this helpful https://www.cs.toronto.edu/~vmnih/docs/Mnih_Volodymyr_PhD_Thesis.pdf

$\endgroup$

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.