cfzd / Ultra-Fast-Lane-Detection

Ultra Fast Structure-aware Deep Lane Detection (ECCV 2020)
MIT License
1.83k stars 494 forks source link

Culane和TUSimple在generate_lines()时逻辑不一致? #346

Open Goooyi opened 1 year ago

Goooyi commented 1 year ago

Culane的generate_line()会像论文中说的,按照k-th location来做expectation of predictions,详见line 24, line 26和line 44, 其中line 44有-1是因为画图的时候图像坐标从0开始对吗? https://github.com/cfzd/Ultra-Fast-Lane-Detection/blob/353df107756b8c03c22c27201e33fc63d84ecfe6/evaluation/eval_wrapper.py#L9-L45

但是在TUSimple的generate_lines()函数中,见line 72, 74, 81: https://github.com/cfzd/Ultra-Fast-Lane-Detection/blob/353df107756b8c03c22c27201e33fc63d84ecfe6/evaluation/eval_wrapper.py#L64-L83 首先72行没像culane一样+1计算出k-th location, 相当于用了每个grid格子的左端点? 然后81行+0.5是因为由于tusimple的griding_num=100所以+0.5相当于从格子左端点变成了格子的中间点?那为什么要/(griding_num-1)来除以区间个数呢?

hwang12345 commented 1 year ago

@Goooyi 有同样的疑问,为什么两个数据集处理方式不同