Sliding windows detection

Why do we have to use 3 sizes of a sliding window, each one bigger than the previous one?

Hi @Doron_Modan

That’s because we didn’t Know the size of each object in all images or in the other words the size of object can be different on different images so first we try to make an different sliding windows with different sizes from small size to large in the lecture prof decide to choose 3 the different sliding windows starting from small window to large window and calculate the accuracy or loss for each window and choose the window which has the high accuracy or low loss .

Note You have the freedom to choose the number of sliding windows you want to start model with. I think 3 was chosen in the lecture for convenience


1 Like

Reasonable explanation, thank you!

@AbdElRhaman_Fakhry It seems like convolutional implementation of sliding windows does not use different size windows. The window size is always the size of images the CNN was trained on (14x14 in the lecture example).

Can You give me the lecture number?
Also The window size is the kernel size not the size of images

Thanks so much for your reply @AbdElRhaman_Fakhry.

I created a post here - Window size in convolutional implementation of sliding windows.

Any feedback would be appreciated.