I am curious why, when applying the sliding window technique, we start out with a small window then gradually increase the window size. Why don’t we start with a bigger window, then if there is an object in that window, just use smaller windows inside of that to pinpoint the object? This sounds more efficient.
In which course are you enrolled?
There are many variants of object detection algorithms, even within the sliding windows family. This video covers one of them, called R-CNN.
You can find other related information using the search term ‘region proposals’. As in all things engineering, there are strengths and weaknesses in each approach.
1 Like