Closed psandovalsegura closed 7 months ago
In ARTrackV2, there is a solution named confidence token. But in ARTrack, i think one effective way is to average the probability of four coordinates we predict in current frame. After that, you can set a overlap to filterate low score predict and consider it as that the target exit the field of vision.
I have video sequences where objects come in and out of frame. If I include every frame of the video in the
Sequence
object, ARTrack predicts a bounding box for every frame. Can ARTrack predict when the template object is not present? How would I implement that?