This question is more about bipartite matching than RT-DETR itself, but I hope it's not an issue.
RT-DETR uses bipartite matching, and from what I understand (please correct me if I'm wrong), bipartite matching issues one-to-one object detection which reduces overlapping by assigning one object to a portion of an image marked by the object's bounding box.
Does this mean RT-DETR will have issues in detecting two different objects with different classes which overlap?
For example, an RT-DETR model is trained to detect helmet and person objects. If the model is used to predict a picture of a person who's holding a helmet in front of their chest (which result in helmet overlapping with person). Will this trouble the model in detecting both helmet and person?
This question is more about bipartite matching than RT-DETR itself, but I hope it's not an issue.
RT-DETR uses bipartite matching, and from what I understand (please correct me if I'm wrong), bipartite matching issues one-to-one object detection which reduces overlapping by assigning one object to a portion of an image marked by the object's bounding box.
Does this mean RT-DETR will have issues in detecting two different objects with different classes which overlap?
For example, an RT-DETR model is trained to detect helmet and person objects. If the model is used to predict a picture of a person who's holding a helmet in front of their chest (which result in helmet overlapping with person). Will this trouble the model in detecting both helmet and person?
Thank you.