0
$\begingroup$

I have gone through various articles on medium and also some from other sites trying to understand SSD. I am able to figure out most of the things from articles except this one.

They always say that the area and location of detectors/anchors in fixed in the model. They also show us the various shapes of detectors based on aspect ratio. But I can't understand how to fix the location of detectors.

Suppose, I want my detector to focus only on the top-right of the image, how can I achieve that? I mean, the filters learned by CNN are applied to the whole image by sliding through its parts. How can we actually achieve this?

Is this related to 'receptive field' or '1x1 convolution'? I am trying to understand these two concepts too but haven't progressed much.

$\endgroup$
  • $\begingroup$ What do you mean by "focus on"? Do you mean "ignore the rest of the image"? $\endgroup$ – D.W. Feb 14 at 0:17
  • $\begingroup$ Yes, how can we make a detector ignore the rest of the image and focus only on the object? $\endgroup$ – Kadam Parikh Feb 14 at 14:06
0
$\begingroup$

If you want to only look for matches in part of the image and ignore the rest, crop out the part you want to look in (removing the rest of the image), then apply the object detector to the cropped image.

$\endgroup$

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.