I have a fully convolutional network that has been trained to classify cats and non-cats in small images (48x48). Because it is fully convolutional, I would expect that if I run it in bigger images, the result will be a classification map with overlapping 48x48 patches where I can predict bounding boxes. However, I am not sure how to convert this map (and the bounding boxes) to the original input dimensions.

I have tried something along the lines of: $$ x_i = (\left \lfloor{x_{c,1} * R}\right \rfloor, \left \lceil{x_{c,2} * R}\right \rceil) $$ where $x_{c}$ represents the x-coordinates of the bounding box in the classification map and $x_i$, in the input image space. $R$ is the ratio between the sizes of the classification map and input image. This doesn't work.

Can someone please explain how to do it? Thanks!


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.