To get confidence score parameter in the sort output

Question

To get confidence score parameter in the sort output

hlmhlr opened this issue 3 years ago · comments

Hi,
Thanks for your repo with multi-class sort. Currently, sort outputs the following parameters:
"frame_num, bbox_x1, bbox_y1, bbox_x2, bbox_y2, category, u_overdot, v_overdot, s_overdot, and identity".

How can I add confidence score value for the tracked objects? The output required is:
"frame_num, bbox_x1, bbox_y1, bbox_x2, bbox_y2, category, u_overdot, v_overdot, s_overdot, identity, and confidence_score".

Your help would be highly appreciated.
Thanks,

Jason (Janghyup) Sohn · Answer 1 · Sat Jan 29 2022 13:23:15 GMT+0800 (China Standard Time)

Hi, do you still need assistance with this issue?

RUIXUEZHAOFENGNIAN · Answer 2 · Fri May 20 2022 18:06:32 GMT+0800 (China Standard Time)

I have the same needs

Jason (Janghyup) Sohn · Answer 3 · Mon May 23 2022 16:51:42 GMT+0800 (China Standard Time)

Hi @RUIXUEZHAOFENGNIAN , I won't be making that modification because it is a breaking change, but I can tell you how to do it yourself:

In sort.py's KalmanBoxTracker class, find __init__() and update(). Put self.conf = bbox[4] in each.
In the same class, under get_state(), add the confidence to the returned tensor with: arr_conf = np.expand_dims(np.array([self.conf]), 0). Put arr_conf in the final np.concatenate function, wherever you'd like.
Make corresponding changes the code that follows.