ContrastiveLosses4VRD
ContrastiveLosses4VRD copied to clipboard
Format of relation annotations?
I wonder how you defined the bbox for the subjects and objects in rel_annotations_train.json?
Do you have any documentation or code on how you built rel_annotations_train.json?
Anyway, I tracked the code to this line(json_dataset_rel.py):
sbj_gt_box = box_utils_rel.y1y2x1x2_to_x1y1x2y2(rel['subject']['bbox'])
are you using y1y2x1x2 format for the annotations?