search

YueLiao / CDN

Code for "Mining the Benefits of Two-stage and One-stage HOI Detection"
Apache License 2.0
89 stars 15 forks source link

some code questions #19

Closed mytk2012 closed 2 years ago

mytk2012 commented 2 years ago

for verb_score in verbscores: #64 query corresponding to verb class_

verb_query = np.sum(np.sort(verb_score.numpy())[::-1][:topN])

        verb_query = torch.max(verb_score)

#find the highest value corresponding to every query obj_scores[index] = verb_query #total 9 verb query put to the object_score query index += 1 thres = np.sort(obj_scores.numpy())[::-1][topN] keep = obj_scores > thres print(keep) out_sub_boxes = outputs['pred_sub_boxes'] out_obj_boxes = outputs['pred_obj_boxes'] What does keep = obj_scores > thres mean? And are my annotations right?

YueLiao commented 2 years ago

Hi, the mentioned codes are for visualization, where we only select the human-object pairs with top-3 confidences for visualization. Your annotations are right.