22bb543b251c39ccdad8063d486987bb-AuthorFeedback.pdf

Infuturework,we16 will investigate other tasks that use response maps, such as human17 joint detection and visual object tracking. R3Q1: Does not handle missing or duplicate annotations.A: Missing annotations are handled by the background37 model (L197-203), which adds a"virtual dot" close to each pixel (equivalent to ahypothetical missing annotation).38