1 | initial version |
I think training a head detector in all directions (front, two sides, back, up-down) together is not the solution, as current solutions usually use only one direction (front, back). I am try to train them separately, and later on, add their detecting result together... any other ideas @StevenPuttemans? Thanks!