How to avoid one object is detected to belong to different categories and the bboxes have high IoU ?

[sisrfeng]

Hi, I set the --vis_thresh 0.2 to get more TP though there will be more FP at the same time.
Then the problem arises:
one object is detected to belong to different categories and the bboxes have high IoU.

If I concatenat results[j] for j in categories __list to get results (a dict as written in ctdet.py) , then the variable results is a list ranther a dict.

Would you mind giving me some guide?
Many thanks !

[sisrfeng]

I am so silly that I expected the concatenated numpy array to be a dict!