Exemplar initialization should have a dataset-dependent part which creates a sequence of (I,bb,name) triplets, and an algorithm-dependent part which uses a specific exemplar framing algorithm.
The process of “framing” an exemplar is choosing a HOG feature size/location given an input image I and a ground-truth bounding box bb.
the full HOG feature pyramid is computed and a slice is selected from the pyramid. For each exemplar, maintain both the slice-bb as well as the “ground-truth” gt-bb. During detection we got a HOG template match at location detection-bb, then estimate the transformation between slice-bb and detection-bb and apply that transformation to gt-bb to get a better localization of the bounding box.