That is actually fairly simple.
- Each stage has to be able to classify the positives, and partially the negatives, to get a weak classifier.
- When no stage has been created yet, finding those negatives is easy, simply sliding window through your larger negative images on different scales.
- However, starting at stage0, when reaching stage1, the positives stay the same, but for the negatives, the misclassified samples of the previous stage are taken+filled with new negatives, that are tested versus stage0.
- This principle continues, so basically at stage10, you are looking for negatives, that are still classified as positive but the previous 10 stages (again we start at 0) ...
So yes it increments in time with each stage.
