clr_head.py 389 line meaning?

[hx121071]

hi, It's a great work. But when I understand the source code, I'm confused with the meaning of "target_yxtl[:, -1] -= (predictions_starts - target_starts)" in clr_head.py 389 line. Can you help me?

[MySuperSoul]

hi, It's a great work. But when I understand the source code, I'm confused with the meaning of "target_yxtl[:, -1] -= (predictions_starts - target_starts)" in clr_head.py 389 line. Can you help me?

The last index of target_yxtl is the length of lane. This is used to compensate the diff between predictions_starts and target_starts since the endpoint of prior and ground truth should be the same.

For example: Assume we have one ground truth lane, the target_start_y is 0, end_y is 20, so the target_length = 20. Then we have one predicted lane prior which pred_start_y is 5, we keep the same end_point so the end_y is also 20, so the pred_length = 15, this is why we need thistarget_yxtl[:, -1] -= (predictions_starts - target_starts).