Thank you for your nice work!
My first question indicates that YOLOx offers positional information and confidence scores for object detection in each frame but doesn't provide class information for each detected bounding box. Traditional SORT/Deep-SORT algorithms usually assume objects of the same class in each tracking and prediction round. If there are multiple classes in the current frame, multiple rounds of matching are carried out. Could you clarify whether class information is not needed in your work?
The second question is whether utilizing YOLOv8 for object proposal would yield comparable performance.
Thank you for your nice work! My first question indicates that YOLOx offers positional information and confidence scores for object detection in each frame but doesn't provide class information for each detected bounding box. Traditional SORT/Deep-SORT algorithms usually assume objects of the same class in each tracking and prediction round. If there are multiple classes in the current frame, multiple rounds of matching are carried out. Could you clarify whether class information is not needed in your work?
The second question is whether utilizing YOLOv8 for object proposal would yield comparable performance.