Closed JacobYuan7 closed 3 years ago
In fact, only one class (index 1), it stands for the confidence of human box. For HICO-DET or V-COCO dataset, it can be ignored. But in real world application, it can be used to suppress a large number of detection-caused false positives by thresholding.
I don't quite understand the self.human_cls_embed in class HoiTR. Why do we need to classify the human box? What do the 2 classes stand for?