The static detector is quite expensive to be called on a full frame, which is why the dynamic detector was introduced in the first place.
The dynamic detector shouldn't start full static detection on the first frame where it loses tracking, it should wait to see if it can recover in the next or few upcoming frames - or else the video processing will lag all of a sudden because of the static's detector slow performances.
This issue relates a bit to #20 - because, with more training images, we also expect to improve the performances as misses will occur less often.
The static detector is quite expensive to be called on a full frame, which is why the dynamic detector was introduced in the first place.
The dynamic detector shouldn't start full static detection on the first frame where it loses tracking, it should wait to see if it can recover in the next or few upcoming frames - or else the video processing will lag all of a sudden because of the static's detector slow performances.
This issue relates a bit to #20 - because, with more training images, we also expect to improve the performances as misses will occur less often.